Debt Overhang: Why Recovery from a Financial Crisis Can Be Slow*

By Satyajit Chatterjee

particularly troublesome feature of the most recent recession has been the painfully slow growth in employment during recent recession has been the painfully slow the recovery. In order for employ- A ment growth to accelerate, economists growth in employment during the recovery. For believe that firms need to invest in employment growth to accelerate, economists new productive capacity. This view is believe that firms need to invest in new productive capacity. typically couched in terms of the need to absorb workers formerly employed This view is typically couched in terms of the need to in the sectors that were most adversely reallocate jobs away from crisis-depressed sectors into other affected by the financial crisis — namely, the construction and financial sectors. But doing so requires an expansion in productive sectors — into other sectors of the capacity in those other sectors. Tepid employment growth economy. The reallocation of jobs away from crisis-depressed sectors requires is a sign that this investment in new productive capacity an expansion in productive capacity has not been forthcoming. One reason for the reluctance to in other sectors. Tepid employment growth is a sign that this investment in undertake productive investment following a financial crisis new productive capacity has not been is debt overhang, a situation in which the existence of prior forthcoming. But it is not for a lack of resources. debt acts as a disincentive to new investment. There are Figure 1 displays the profits of the non- other explanations that, to varying degrees, account for the financial corporate sector and shows that profits rose strongly during this re- current reluctance of U.S. corporations to invest. In this covery. And if we examine the disposi- article, Satyajit Chatterjee focuses on the debt overhang tion of investible funds, we discover that the nonfinancial corporate sector problem. has dramatically reduced its investment in productive capacity relative to the resources available for investment. In their widely read book, Car- their work is that economic recovery This is evident in Figure 2, which men Reinhart and Kenneth Rogoff from bad financial crises tends to be shows capital outlays of the nonfinan- have marshaled an impressive amount slow. On average, it takes an economy cial corporate sector as a percentage of of data on global financial crises going somewhere around seven years follow- funds that the nonfinancial corporate back eight centuries. One lesson from ing a crisis to get economic activity sector already possesses (without re- back to its normal trend path. In some course to any new borrowings or equity cases, the return to trend can take Satyajit issues — so-called “internal funds”) Chatterjee is a much longer — close to two decades! and can use for this purpose. This senior economic This historical experience reso- advisor and percentage fell precipitously during nates with our current situation. A economist in the recession and has since remained particularly troublesome feature of the the Philadelphia depressed. These facts indicate that Fed’s Research Department. the U.S. nonfinancial corporate sector This article is * The views expressed here are those of the au- possesses investible resources but has available free thor and do not necessarily represent the views chosen not to deploy these resources in of charge at www.philadelphiafed.org/ of the Federal Reserve Bank of Philadelphia or productive investments during the re- research-and-data/publications/. the Federal Reserve System. www.philadelphiafed.org Business Review Q2 2013 1 covery. Since our current slow recovery FIGURE 1 is partly attributable to the reluctance of businesses to invest in productive Domestic Nonfinancial Corporate assets, we need to understand why Profits Before Tax with IVA and CC financial crises have this effect on Adjustments investment.1 Economists believe that one Billions of USD (SAAR) reason for the reluctance to under- 1,200 take productive investment following a financial crisis is debt overhang. 1,000 Debt overhang is a situation in which the existence of prior debt acts as a 800 disincentive to new investment. When a firm has outstanding debt on which 600 the likelihood of default is significant, any investment that improves the 400 firm’s future profit potential also increases the value of outstanding debt. 200 All else remaining the same, an in- Last quarter plotted: 2012Q3 crease in the value of outstanding debt 0 reduces the value of equity in the firm; 1997 1999 2001 2003 2005 2007 2009 2011 that is, it results in a wealth transfer from equity owners to existing credi- Sources: BEA, Haver tors. Since equity owners are the ones who make investment decisions, the transfer acts like a tax on the return FIGURE 2 on new investment. This “tax” results in a drop in the rate of investment in business capital, which, in turn, slows Nonfinancial Corporations: Capital down the recovery. Outlays/U.S. Internal Funds (SA) There are other possible explanations for the reluctance of U.S. companies to invest. One oft-cited reason Percent of U.S. Internal Funds 160 is “increased uncertainty about the future.” When investment decisions 150 are costly to reverse, there is value 140 in waiting and learning more about 130 future conditions before committing 120 funds to a project. Thus, increased

110 uncertainty about the future may

100

80 1 One might think that the reluctance to add new productive capacity results from current 70 capacity utilization rates being low. If existing Last quarter plotted: 2012Q3 capacity is not being fully utilized, why expand 60 capacity? True, but it raises the question of why 1997 1999 2001 2003 2005 2007 2009 2011 utilization rates are low. If corporations as a whole were investing more, capacity utilization rates would go up right away. One must consider the possibility that low capacity utilization is a Sources: FRB Flow of Funds, Haver symptom of some deeper malady that is affecting investment – not the malady itself.

2 Q2 2013 Business Review www.philadelphiafed.org cause companies to delay investment. sis. The crisis caused the U.S. banking nancial corporate sector scaled by Aside from the increased uncertainty sector to deleverage. In doing so, banks the gross value added in the sector.4 that inevitably accompanies a deep cut off credit to the nonfinancial sector During much of the boom period, the recession, commentators have pointed — the now infamous “credit crunch.” liabilities of the sector shrank relative to uncertainty about the future path Because credit is a fundamental ingre- to its GDP. Nevertheless, it is true of U.S. fiscal (tax and expenditure) dient in the smooth operation of asset that whatever debt there was became policy as well as uncertainty about the markets, the crunch adversely affected much more risky following the onset of impact on businesses’ health-care costs the value of all types of tangible busi- the financial crisis in the fall of 2008. resulting from the recently enacted ness capital. The steep drops in the Figure 4 shows the difference in yields Affordable Care Act as factors holding value of assets owned by the nonfi- on medium-term industrial bonds and back investment and hiring. Another nancial corporate sector also lowered U.S. Treasuries. The difference is the explanation may be that the growth the sector’s net worth and raised the additional return required by investors rate of (multifactor) productivity has frequency of business failures.3 Both to absorb the default risk present in fallen back to its historical norm from factors made corporate debt appear industrial bonds but absent in Treasury the above-average pace experienced more risky to investors. bonds. As one can see, the compensa- during the decade preceding the onset It is worth observing that “exces- tion for default risk (the so-called risk of the financial crisis, causing the sive borrowing” by the nonfinancial spread) rose dramatically as the crisis rate of investment growth to decline corporate sector during the boom years unfolded and remains elevated today. in tandem. Finally, it is thought that is not part of this narrative. Figure Although risk spreads can go up retiring baby boomers may be holding 3 shows the liabilities of the nonfi- back business investment by depressing equity values as they sell stocks to fund their retirement. More fundamentally, 4 Scaling by sector GDP takes account of the 3 The net worth of the nonfinancial corpo- fact that borrowing is a natural complement a more slowly growing labor force re- rate sector is simply the difference between of economic activity and tends to go up with quires less growth in capital equipment the value of its assets and its liabilities. The it. Thus, to determine if the sector indulged in to productively equip new workers mechanism through which a drop in net worth “excessive” borrowing, it is important to look at amplifies a credit crunch is discussed in more its liabilities relative to a measure of economic joining the labor force, so there is less detail in my 2010 Business Review article. activity. growth in investment. Of course, all of these explanations, to varying degrees, FIGURE 3 account for the current reluctance of U.S. corporations to invest. In this Nonfinancial Corporations: Liabilities as a article I focus on the debt overhang problem.2 Share of Gross Value Added

Ratio to Gross Value Added FINANCIAL CRISIS AND THE 2.2 GENESIS OF DEBT OVERHANG The genesis of the debt overhang 2.0 problem lies in the recent financial cri- 1.8

2 Following the onset of the financial crisis, a 1.6 number of researchers and many commentators have pointed to debt overhang as a reason for the drop in investment and its slow recovery. 1.4 The article by Thomas Philippon and the commentary by Filippo Occhino, for instance, discuss the debt overhang problem as it pertains 1.2 to the current crisis. Occhino and Andrea Pescatori’s article discusses the role of debt Last quarter plotted: 2012Q3 1.0 overhang in constraining investment during 1997 1999 2001 2003 2005 2007 2009 2011 business downturns more generally. The article by Karen Croxson, Susan Lund, and Charles Roxburgh stresses the global extent of the debt overhang problem and looks broadly at both Sources: BEA, FRB Flow of Funds, Haver private-sector and public-sector debt. www.philadelphiafed.org Business Review Q2 2013 3 for many reasons, the evidence is sug- gestive of a crisis-induced increase in FIGURE 4 default risk as well as loss rates given default. Figure 5 displays the number Corporate Bond Spreads of business bankruptcy filings. Filings were on an upward trend even before the crisis, but they have accelerated Difference in Yields since the third quarter of 2008. Al- 6 though filings have come down, they remained elevated relative to the boom 5 years until recently. Figure 6 displays the ratio of credit market debt of the 4 nonfinancial corporate sector and the value of tangible assets in this sector. 3 As shown, the ratio rose from around 2 42 percent at the start of 2007 to more than 56 percent at the height of the 1 crisis. The ratio is currently above 50 percent. A higher value of debt Last quarter plotted: 2012Q3 0 relative to tangible assets is a concern 1997 1999 2001 2003 2005 2007 2009 2011 for creditors because tangible assets are what creditors mostly recover if a company fails. A loan-to-value ratio of Note: Current Treasuries 5-10 years subtracted from corporate industrial bond 5-10 years, 50 percent is an indication to creditors yield to maturity that they should now expect higher Sources: Bank of America, Merrill Lynch, Haver loss rates (given default) compared with the pre-crisis years.5 Finally, there is direct evidence FIGURE 5 of a greater likelihood of default or an increase in expected loss rates given Business Bankruptcy Filings default. This evidence comes from credit default swap (CDS) spreads on bonds issued by highly reputable U.S. Filings 18,000 corporations.6 A CDS written on a specific corporate bond is an agree- 16,000 ment in which the seller of the CDS 14,000 promises to compensate the buyer for 12,000

10,000

8,000 5 On the face of it, a loan-to-value ratio of 50 percent suggests that creditors will not take 6,000 any losses in case of bankruptcy. However, the value of the firm’s tangible assets is much lower 4,000 in bankruptcy than its reported value when the firm is a going concern. Indeed, it is not uncom- 2,000 mon for creditors to dispose of recovered assets Last quarter plotted: 2012Q3 0 at huge discounts. These so-called “fire sales” occur because it is costly for creditors to hold on 1997 1999 2001 2003 2005 2007 2009 2011 to recovered assets.

6 The index is based mostly on the corporate debt of nonfinancial firms. The few financial firms that are included in this index are firms Sources: Administrative Office of the U.S. Courts, Haver whose debt maintained a top credit rating through the crisis.

4 Q2 2013 Business Review www.philadelphiafed.org any losses incurred due to default on the named bond. In return, the buyer FIGURE 6 pays the seller an insurance premium each period. This insurance premium Nonfinancial Corporations: Credit is measured as a percent of the face Market Debt/Tangible Assets value of the bond and is referred to as the CDS spread. A high spread means that default on the bond is more likely, Credit Market Debt over Tangible Assets 0.60 that the loss incurred in the event of default is higher, or both. As Figure 7 shows, the CDS spread was around 50 0.56 basis points (a basis point is 1/100 of a percent) prior to the crisis, then rose 0.52 dramatically during the crisis, and is still almost twice as high compared 0.48 with the pre-crisis period. The bottom line is that corporate sector debt began to look substantially 0.44 more risky to investors following the fi- Last quarter plotted: 2012Q3 nancial crisis, mostly because the crisis 0.40 depressed asset values. 1997 1999 2001 2003 2005 2007 2009 2011

DEBT OVERHANG: WHAT IT MEANS AND WHY IT’S Sources: FRB Flow of Funds, Haver BAD NEWS In his 1995 article, Owen Lamont gives an example of what debt over- FIGURE 7 hang means and why it is bad for investment. Suppose that a firm has $100 in debt, due next year, but will CDS Spreads for Investment Grade Bonds have assets worth only $80. Thus, the firm will not have enough resources to Basis Points meet its debt obligations next year and 300 will default for sure. Now suppose that a business opportunity presents itself to 250 this firm in the form of a project that will cost $5 today and yield $15 next 200 year. If existing creditors are first in line for the payout of the firm, no out- 150 side investors will be willing to supply $5 to the firm because the benefit will 100 go to the original creditors, who will have their payoff go up to $95. Lamont 50 calls the $20 gap between assets and liabilities the debt overhang. If the net Last quarter plotted: 2012Q2 0 payoff from the new investment cannot 2004 2005 2006 2007 2008 2009 2010 20122011 cover this gap, the project will never be financed by an outside investor. Debt overhang raises the bar for new investments: Only very profitable Sources: Markit CDX.NA.IG, Bloomberg investments will be worth undertak- www.philadelphiafed.org Business Review Q2 2013 5 ing. In this example, the return on the percent probability that the firm will Empirical estimates of the ef- $5 dollar investment would have to be go bankrupt, but instead of receiving fects of debt overhang for individual at least $25 to make the investment $80, the creditors will get $95 in the corporations appear to be quite large. worthwhile to the outside investor. event of default. Therefore, the market According to the study by Christo- The return would enable the firm to value of existing debt will rise to (½) × pher Hennessey, Amnon Levy, and repay what is owed to the original $95 + (½) × $100 = $97.50. Corre- Toni Whited, a 1 percent increase in creditor and still make a positive re- spondingly, the market value of equity leverage for a corporation with median turn on the investment. will rise to (½) × $0 + (½) $25 = leverage leads to a 1 percent decline in It is easy to generalize this exam- $12.50. The important point to note investment for that corporation. While ple to the case where default is prob- here is that although the total value it is not easy to translate this estimate able but not certain. First, assume that of the firm rises by $15 (the payoff into an estimate of the reduction in ag- if no new investment is undertaken, from the new investment), half of the gregate business fixed investment due the value of assets in the next period overall increase in value goes to cur- to the debt overhang problem, it sug- can be either $80 or $110 with equal rent creditors and half to owners. The gests that the effect is potentially sig- probability. Thus, there is a 50 percent implicit expected percentage of the nificant. In the aggregate, the leverage chance that the firm will be bankrupt “tax” imposed by current creditors on of the nonfinancial corporate sector and the creditors will get $80, and the return on new investment to equity (measured as the ratio of its liabilities there is a 50 percent chance that the firm will not be bankrupt, in which case we may assume that the creditors It is worth pointing out that the debt overhang will receive $100.7 The market value of the firm’s debt is then (½) × $80 + problem can be eliminated if the returns to new (½) × $100 = $90.8 Correspondingly, investment can be dedicated solely to new the market value of the firm’s equity investors. (i.e., the value of the firm to its owners) is (½) × $0 + (½) × $10 = $5 (which follows from the fact that when the firm is bankrupt, the owners lose holders is 50 percent, which is simply to its net worth) is around 13 percent everything, and when it is not bank- the probability of bankruptcy. higher now than before the crisis, sug- rupt, the owners retain the difference The fact that the return to owners gesting that business investment may between the value of the assets and the from undertaking a new investment now be 13 percent lower as a result of value of the liabilities). Now, assume is adjusted downward by the prob- debt overhang. Over a four-year period the new investment is undertaken. ability of default on existing debt is (the third quarter of 2008 to the third Then, the value of the firm’s assets what financial economists call the quarter of 2012), this would amount to 9 in the next period will be either $95 “debt overhang” problem. Simply put, annual growth in business investment (which is the sum of $80 plus $15, the in the event of default, the returns to that is about 2 percent slower than latter being the return from the new any new investment will first accrue to what it would have been had the crisis 10 investment) or $125 (which is the sum the creditors rather than to the equity not intervened. of $110 and $15). Notice that even holders, and this fact lowers the return It is worth pointing out that the with the new investment, there is a 50 to equity holders from funding new debt overhang problem can be elimi- investment projects. All else remaining the same, the overhang can be expected to reduce investment by lev-

7 In this eventuality, the firm can borrow eraged corporations. Said differently, 10 Normally, a lower level of business fixed $100 again from the same or a different set of the debt overhang raises the required investment can be expected to be partially creditors and pay off the loan that has come rate of return for new investment to be offset by an increase in some other component due. The process of using new loans to pay off of aggregate demand (such as higher consumer maturing debt is called “rolling over” the debt. undertaken. spending on durables), and the overall effect on real GDP would be smaller than that implied by 8 For simplicity, I have assumed that the interest a 13 percent decrease in business fixed invest- rate on safe financial investments (say, a one- ment alone. But when there is slack in resource year Treasury bond) is zero. If the interest rate utilization (as evidenced by the current high were positive, say, 1 percent, the market value of 9 See, for instance, the articles by Christopher unemployment and low capacity utilization the firm’s debt would be $90 ÷ 1.01. Hennessey and Stewart Myers. rates), there may not be any offset.

6 Q2 2013 Business Review www.philadelphiafed.org nated if the returns to new investment investors would be unwilling to invest On average, the new project will fetch can be dedicated solely to new inves- in new projects unless these projects an additional $7.50 tomorrow. This tors. This is not possible if new inves- are very profitable. Consequently, the amounts to an expected rate of return tors are given equity shares in the firm rate of growth of business investment of (7.50 − 5)/5 × 100 = 50 percent. because, by law, equity holders cannot is adversely affected by the presence of This might look like an attractive be paid off unless all creditors are paid risky debt. return, except that when default is off first. In other words, creditors have a possibility, there might be another a senior claim on the income and assets DEBT OVERHANG AND strategy that will fetch the owners an of the firm vis-à-vis equity holders. On THE INCREASED VALUE OF even more attractive return. the other hand, if the new investment LIQUIDITY Suppose that the firm’s owners is debt financed (i.e., the firm issues So far, I have considered the can keep the $5 in the firm as cash debt rather than equity to its new incentives of outside investors (equity and, in the next period, decide if investors), the debt overhang problem holders or new creditors) to invest in they want to pursue the new invest- boils down to whether new creditors a new project. However, as we have ment after learning about the value have a senior claim to the income and seen, the nonfinancial corporate sec- of their existing assets. The returns assets of the firm vis-à-vis existing tor is not starved for funds. For debt from this strategy are displayed in the creditors. If they do, the debt overhang overhang to be an explanation for bottom row of Table 1. If the value of problem again disappears.11 In prac- lackluster investment, we also need to the assets turns out to be $80 (which tice, creditors typically insist that their consider the firm’s incentives to invest happens with a 50 percent probability), claims be senior to the claims of any its own funds in the new project. they have $85 on hand. Since they future creditor of the firm so that the I will do this by going back to owe $100, they are bankrupt. At this debt overhang problem remains even if the example where the future value point, suppose they are able to take $1 the new investment is debt financed.12 of the firm’s assets is uncertain (and out of the $5 as profits and hand the The bottom line is that if a firm can be either $80 or $110). Imagine firm over to the creditors.13 So, with has debt outstanding on which there is now that the $5 is actually the firm’s a 50 percent probability, the owners a positive probability of default (risky own money, obtained as profits from debt), the presence of that debt low- current operations. What should the ers the returns to equity owners from firm do with it? The top row of Table 13 Bankruptcy law makes it illegal for corporations to distribute any dividends in a state of new investment. This is because in 1 shows what the firm can get if it insolvency. Thus, the example is not to be the event of default, all of this new invests its $5 in the new project today. taken literally. Rather, it is intended to capture investment is lost. In this situation, With a 50 percent probability, the firm the fact that owners do have opportunities to legally take money out of the firm when bank- will go bankrupt and all of the return ruptcy is probable but not certain. The assump- from the project will be lost, and with tion that only a portion of total cash holdings can be taken out in this manner acknowledges a 50 percent probability, the firm will the limitations that exist on this type of equity 11 For instance, in the example, suppose that all survive and the project will return $15. extraction. of the new investment is financed by new debt. Since the new investment costs $5, the firm will owe $105 next period. The probability of default is still 50 percent, since it will occur only if the value of assets turns out to be $95. But if the $5 claim of the new investors is senior to the $100 TABLE 1 claim of existing creditors, new creditors can be paid off even in bankruptcy because the value of the firm’s assets ($95) is sufficient to cover the Payoff in Payoff Outside Average $5 claim of new creditors. Given this, new credi- Bankruptcy of Bankruptcy Return tors would view the loan as a safe investment and would presumably go ahead and finance Investment (50 percent (50 percent Average (Average Payoff the investment project. In contrast, if the claim Strategy chance) chance) Payoff − 5)/5 * 100 of new creditors is junior to the claims of exist- Invest $5 Now $0 $15.00 $7.50 50 percent ing creditors, they get nothing in the event of default because the $100 claim of existing credi- Hold $5 in $1.00 $15.00 $8.00 60 percent tors will exhaust all of the firm’s assets. Cash & Invest Tomorrow If 12 It would take us too far afield to fully explain Not Bankrupt the reasons why existing creditors insist on the seniority of their claims vis-à-vis future creditors. The article by Burcu Eyigungor sheds light on this issue. www.philadelphiafed.org Business Review Q2 2013 7 get back $1. If the value turns out to To understand his point in the pour new money into the firm. Also, be $110, they have $115 on hand, and context of our example, suppose that while the payoff from the “hold on to their assets are worth more than their the business sector’s collective reluc- cash” option declines to $6.60 and its liabilities. At this point, they can ask tance to invest increases the prob- rate of return to 32 percent, the dif- their creditors to roll over the $100 ability of the bad outcome (low asset ference in the rate of return between debt and invest the $5 in the new value) from 50 percent to 60 percent. the two strategies widens to 12 percent investment project and earn $15 in the Now the “tax” on new investment is 60 from 10 percent. Thus, the strategy of following period. So, with a 50 percent percent, and as shown in the top row just hanging on to the cash will seem probability, the owners get back $15. of Table 2, the average payoff from in- even more attractive to business own- The expected payoff from just hang- vesting $5 today declines to $6 and the ers. ing on to the $5 as cash today is then average return declines to 20 percent. In sum, an increase in pessimism (½) × $1 + (½) × $15 = $8 and the The decline in the rate of return would (by which we mean a greater prob- expected return is (8 − 5)/5 × 100 = make outside investors (be they equity ability weight on the bad outcome) 60 percent. Since 60 percent beats 50 owners or creditors) more reluctant to makes the “tax” imposed by the debt percent, the firm’s owners are likely to be tempted to just keep their profits as FIGURE 8 cash in the firm and decide what to do with it in the next period. The bottom line is that cash Nonfinancial Corporations: Financial Assets has the benefit of liquidity: It gives as Share of Gross Value Added equity owners the option to take some of their money out if bankruptcy Ratio to Gross Value Added becomes more probable. Thus, when 2.2 there is a relatively high probability 2.0 of bankruptcy, equity owners have an incentive to delay making real 1.8 investments and accumulate cash with the intention of taking that cash 1.6 out as dividends at some point in the 1.4 future. This seems consistent with the evidence. As shown in Figure 8, the 1.2 ratio of financial assets to gross value Last quarter plotted: 2012Q1 added in the nonfinancial corporate 1.0 1997 1999 2001 2003 2005 2007 2009 2011 sector has risen during this recovery.

DEBT OVERHANG AND Sources: BEA, FRB Flow of Funds, Haver SELF-FULFILLING PESSIMISM Many current observers of the U.S. economy hold the view that for TABLE 2 an economy growing slowly from a depressed state, it does not take much Payoff in Payoff in in terms of some adverse shock to Bankruptcy Bankruptcy Average Return tip it into a recession. This being the Investment (60 percent (40 percent Average (Average Payoff case, our current slow recovery has Strategy chance) chance) Payoff − 5)/5 * 100 engendered greater pessimism about Invest $5 Now $0 $15.00 $6.00 20 percent the economy’s future growth prospects. Hold $5 in $1.00 $15.00 $6.60 32 percent An important point that Lamont Cash & Invest Tomorrow If makes in his article is that in the Not Bankrupt presence of a debt overhang problem, pessimism about the future can be self- perpetuating.

8 Q2 2013 Business Review www.philadelphiafed.org overhang problem higher and retards carrying debt, the loss of net worth to liquidate when business conditions business investment even more. Slow brings them closer to default. Debt deteriorate and bankruptcy becomes growth in business investment, in turn, overhang occurs when there is a more likely. On both counts, the can keep a lid on the speed of econom- significant probability that a firm will rate of investment in business capital ic recovery and makes pessimism about go bankrupt in the near future. The is adversely affected. Thus, debt the future self-perpetuating. overhang of existing debt reduces overhang is one potential explanation the incentives of new investors to for why firms have been reluctant to CONCLUSION invest in business capital because, expand capacity in this recovery. The Recovery from financial crises in the event of default, part of the macroeconomic consequence of this tends to be slow, and one reason for return on new investment accrues reluctance to invest is a slow recovery. this is the debt overhang problem. to existing creditors. Debt overhang To the extent that a slow recovery The declines in asset values that also increases owners’ incentives to engenders pessimism, it exacerbates accompany a financial crisis lower invest their current profits in financial the debt overhang problem. firms’ net worth. If these firms are assets because these assets are easier

REFERENCES

Chatterjee, Satyajit. “De-Leveraging and Hennessey, Christopher A., Amnon Levy, Occhino, Filippo, and Andrea Pescatori. the Financial Accelerator: How Wall and Toni M. Whited. “Testing Q Theory “Debt Overhang in a Business Cycle Mod- Street Can Shock Main Street,” Federal with Financing Frictions,” Journal of Finan- el,” Federal Reserve Bank of Cleveland Reserve Bank of Philadelphia Business Re- cial Economics, 83 (2007). Working Paper 10-03R (December 2010). view (Second Quarter 2010). Lamont, Owen. “Corporate Debt Over- Philippon, Thomas. “The Macroeconom- Croxson, Karen, Susan Lund, and Charles hang and Macroeconomic Expectations,” ics of Debt Overhang,” paper presented at Roxburgh. “Working Out of Debt,” McKin- American Economic Review, 85:5 (Decem- the 10th Jacques Polak Annual Research sey Quarterly (January 2012). ber 1995). Conference, Washington D.C., November 5-6, 2009. Eyigungor, Burcu. “Debt Dilution: When It Myers, Stewart, C. “Determinants of Is a Major Problem and How to Deal with Corporate Borrowing,” Journal of Financial Reinhart, Carmen, and Kenneth Rogoff. It,” Federal Reserve Bank of Philadelphia Economics, 5:2 (October 1977). This Time Is Different: Eight Centuries of Business Review (forthcoming). Financial Folly. Princeton, NJ: Princeton Occhino, Filippo. “Is Debt Overhang University Press, 2009. Hennessey, Christopher. “Tobin’s Q, Debt Causing Firms to Underinvest?” Federal Overhang and Investment,” Journal of Fi- Reserve Bank of Cleveland Economic Com- nance, 59:4 (August 2004). mentary, 2010-7 (July 2010).

www.philadelphiafed.org Business Review Q2 2013 9 DSGE Models and Their Use in Monetary Policy*

by Michael Dotsey he past 10 years or so have seen the conjunction with other tools com- development of a new class of models that are monly employed by monetary poli- proving useful for monetary policy: dynamic cymakers. These other tools include T purely statistical models, often not tied stochastic general equilibrium (DSGE) to any particular economic theory, but models. Many central banks around the world, including instead are solely based on historical regularities found in the data. Such the Swedish central bank, the European Central Bank, tools also include large macroeconomic the Norwegian central bank, and the Federal Reserve, models that contain many sectors of the economy but generally do not place use these models in formulating monetary policy. In many theoretical restrictions on the this article, Mike Dotsey discusses the major features interrelationships between the various economic sectors. Other tools include of DSGE models and why these models are useful to economic surveys of consumers, firms, monetary policymakers. He outlines the general way or forecasters, as well as policymakers’ own expertise. in which they are used in conjunction with other tools These other tools provide valuable commonly employed by monetary policymakers and insights into the state of the economy that complement the insights derived points out the promise of using these models as well as from explicit theoretical models, which the pitfalls. account for important interactions between sectors of the economy. Togeth- er, the various modeling approaches The past 10 years or so have wit- steps, a number of other central banks comprise the toolkit that policymakers nessed the development of a new class have incorporated DSGE models into commonly rely on. This article will of models that are proving useful for the monetary policy process, among concentrate on DSGE models, which monetary policy: dynamic stochastic them the European Central Bank, share the strengths of many theoreti- general equilibrium (DSGE) models. the Norge Bank (Norwegian central cally grounded models but are designed The pioneering central bank, in terms bank), and the Federal Reserve.2 with the intention of providing fore- of using these models in the formula- This article will discuss the major casts and identifying the key drivers tion of monetary policy, is the Sveriges features of DSGE models and why of current economic activity. In doing Riksbank, the central bank of Swe- these models are useful to monetary so, I will point out the promise of this den.1 Following in the Riksbank’s foot- policymakers. It will indicate the modeling strategy as well as its pitfalls. general way in which they are used in Economic models, in general, Mike Dotsey is provide valuable guidance when for- a vice president 1 See the article by Malin Adolfson and coau- mulating monetary policy. Because and senior thors. the economy is so complex and key economic policy economic components are intertwined, advisor in the 2 Examples of these models can be found in Philadelphia Smets and coauthors; Bruback and Sveen; and it is necessary to develop frameworks Fed’s Research Chung, Kiley, and Laforte. that capture these interrelationships. Department. In order to capture, say, the effect that This article is *The views expressed here are those of the an increase in productivity has on available free author and do not necessarily represent of charge at www.philadelphiafed.org/ the views of the Federal Reserve Bank of consumption, we must have a model research-and-data/publications/. Philadelphia or the Federal Reserve System. that incorporates the behavior of many

10 Q2 2013 Business Review www.philadelphiafed.org variables, such as income, investment, tors of production, such as labor and structure imply that each of these types labor supply, and consumption, if we capital, in ways that minimize their of shocks has very different implica- are to understand this effect. Simply costs. This depiction of behavior places tions for the economic predictions of looking at one equation that attempts restrictions on the actions of firms, the model, and the estimation of the to only model consumption is likely to households, and the government in the model places weights on each type of produce an incomplete and mislead- model, and the validity of these restric- disturbance that allows the model to fit ing interpretation. Thus, a model that tions can be formally tested. Doing so the data as best as possible. integrates many economic components allows model builders a way of analyz- Finally, the models are inherently is necessary for understanding and ing the strengths and weaknesses of dynamic. Current behavior does not predicting economic behavior. the underlying theory. Model restric- depend only on the current economic However, because all models are tions that are not consistent with climate but also on anticipation of approximations of actual economic economic data indicate a weakness what the future holds. For example, behavior, it is often useful to combine that calls for further development of firms’ hiring and investment deci- the insights from a number of models the model. When various restrictions sions depend on whether they believe along with statistical forecasts and the are consistent with the data, we can that economic demand will be weak individual experience of policymakers. have more confidence in the model. It or strong in the future, not just on That is generally what many central is safe to say that no model has been current demand conditions. This dy- banks do, and DSGE models are in- developed that is consistent with all namism implies that expectations of creasingly becoming a part of policymakers’ toolkits. DSGE models are small to medium size

AN OVERVIEW economic models that incorporate the major OF DSGE MODELS sectors of the economy into a coherent and DSGE models are small to medium size economic models that incorpo- interrelated whole. rate the major sectors of the economy into a coherent and interrelated whole. features of the actual economy, but the future play an important role, and They are general equilibrium in na- great strides have been made, and the although such an assumption is not ture, meaning that prices and interest underlying methodology incorporated required, most DSGE models assume rates adjust until supply equals demand into the development of these models that the actors in the model — indi- in every market. In particular, the makes further improvements likely. viduals and firms — form expectations demand for goods equals the supply of The models are also stochastic, that are consistent with the underlying goods, the demand for assets equals meaning that they incorporate the theoretical framework of the model. the supply of assets, and the demand random components that play an im- This does not imply that households for labor equals the supply of labor. portant role in explaining the cyclical and firms perfectly anticipate future Further, these models include a behavior of the economy. Common dis- outcomes but that, on average, they do private sector composed of households turbances include shocks that change not make systematic errors. This type and firms, as well as a public sector consumer demand, shocks that influ- of expectations formation is referred made up of a government fiscal author- ence the behavior of financial markets, to as “rational expectations,” and it ity and a central bank. A distinguish- and changes in economic productivity is a common feature of a broad set of ing feature of these models is that con- that affect the efficiency of production. economic models. sumers and firms in the model make What is key to the DSGE paradigm Combining these ingredients — decisions that maximize welfare and is that these shocks can be estimated the use of explicit maximizing behav- profits, respectively. Individuals make as can the proportions of changes in ior that is also dynamic in nature and decisions about consumption and labor economic activity that are due to a forward-looking rational expectations supply that maximize their economic particular disturbance. For instance, — makes the output of DSGE models, well-being subject to constraints based we may ask what part of the latest whether that output is an economic on their wealth. For instance, individu- recession was due to financial shocks forecast, the results of a policy experi- als in the model cannot consume more as opposed to changes in productivity ment, or the analysis of the sources of than they can afford. Firms set prices or fiscal policy shocks. The restric- economic fluctuations, readily inter- that maximize profits and demand fac- tions imposed on the model’s economic pretable in terms of economic theory. www.philadelphiafed.org Business Review Q2 2013 11 Thus, DSGE models paint a coherent prices is helpful information in esti- erty for policymakers to understand in picture with respect to a host of issues mating the price-setting parameters of using economic models for informing that are of interest to policymakers. the typical DSGE model. particular policy actions. Estimation also pays dividends. Therefore, it is useful to look at MAKING THE MODELS One outgrowth of statistical estima- the implications of a number of models OPERATIONAL tion is that it allows us to character- in order to compare the performance All of the relationships that ize the data uncertainty surrounding of different theories and evaluate govern the economic behavior of any the parameter estimates. Are we fairly which particular ways of thinking DSGE model include parameters, and certain of a given parameter’s value, about the economy lead to a bet- these parameters must be assigned values before the model can be used. For Using a number of different models allows instance, we need to know how much individuals value current consump- economists and policymakers to ascertain the tion relative to future consumption in extent of model uncertainty, which involves the order to understand their consumption and saving decisions. The parameter uncertainty that arises because all economic that governs that aspect of behavior is models are approximations of behavior, and called a discount factor, and it must be given a specific value. Also, we need to no model accurately captures all facets of understand the costs associated with economic activity. a firm’s adjustment of its capital stock if we are to understand investment be- or could that parameter take values ter understanding of actual behavior. havior, and there are parameters that that span a wide range? The esti- Thus, examining model uncertainty govern the magnitude of these costs. mation also allows us to capture the is an important part of analyzing the They too must be either calibrated or uncertainty surrounding the economic output of DSGE exercises, since like all estimated. Generally, the models are forecasts, as well as the uncertainty economic models, DSGE models are, estimated using historical data because surrounding the results regarding the to some extent, misspecified. Compar- it is not obvious what the appropriate likely consequences of using an alter- ing the output of many DSGE models values of many of the parameters are. native monetary policy. sheds light on the confidence we have Furthermore, estimation allows us to Further, using a number of dif- in any particular implication of the establish the uncertainty surrounding ferent models allows economists and models as a whole. Hence, looking at any particular parameter value. That, policymakers to ascertain the extent a number of different models helps in turn, allows us to better under- of model uncertainty, which involves policymakers assess the risk of any par- stand the uncertainty inherent in the the uncertainty that arises because all ticular viewpoint based on a particular predictions of the model. Thus, all the economic models are approximations model. As indicated in the June 2011 mathematical relationships that govern of behavior, and no model accurately minutes of the Federal Open Market the economic behavior of any DSGE captures all facets of economic activ- Committee meeting, DSGE models are model include parameters that require ity. Thus, different models analyz- being studied by staff members at the estimation. ing the same question will come up Board of Governors and at the Federal Usually, the estimation is done with different implications, and as a Reserve Banks of Chicago, New York, using a methodology called Bayesian result, there is uncertainty about those and Philadelphia. If models that differ statistics, which allows the user to in- implications. Along with this type of along various dimensions all point to corporate prior knowledge of the econ- uncertainty, there is uncertainty that the same conclusion, the policymaker omy. For example, this information characterizes each particular model can be more reassured about the out- may come from microeconomic studies because the parameters of each model come of a particular decision. and thus may contain information that are estimated and not known exactly. is not formally part of the model but Economists are, in general, more A MORE DETAILED DEPICTION is nonetheless useful for gauging the uncertain about their models than OF A BASIC MODEL likely value of the model’s parameters. they are about the parameters of any The structure of a basic DSGE, For example, microeconomic evidence particular model, making the degree of namely, the model developed by staff on how frequently firms adjust their model uncertainty an important prop- members at the Federal Reserve Bank

12 Q2 2013 Business Review www.philadelphiafed.org of Philadelphia, is displayed in the fig- power, and they set prices in order to and workers produce the same amount ure.3 The model is nicknamed PRISM, maximize profits over time. The price of output — it is questionable whether which stands for Philadelphia Research of each good is adjusted at randomly the price-setting mechanism enjoys Intertemporal Stochastic Model. As is selected intervals, with only a subset that property. For example, as inflation true of much of the DSGE modeling of firms adjusting their prices at any changes, we would expect the fre- framework, the foundations are based point in time.4 Thus, the price level is quency with which prices are changed on New Keynesian economics, which sticky, which means that it does not to vary as well, but this behavior is not explicitly models various forms of price adjust instantaneously to economic part of the theoretical pricing mecha- and wage rigidity thought to be an disturbances. The particular pricing nism in the model. integral part of a modern economy’s behavior that maximizes economic Along with a productivity shock, structure. The firms in PRISM employ profits over time is one in which firms firms’ decisions are influenced by workers and rent capital in order to reset their prices as a markup over a shocks to the markup of price over produce goods, and they do so in a weighted average of current and future marginal cost. We may think of this manner that minimizes the cost of marginal costs. Price rigidities are an type of shock as a random variation in producing output. Production is also important feature of the model and are a firm’s market power, perhaps influ- subject to productivity shocks. Firms an important element in aligning the enced by the random inflow and out- also enjoy some monopoly or pricing model with the data. flow of the number of competing firms. While the production function, Households in the model own which indicates the amount of output the firms and the capital stock. They 3 The features described are fairly similar across that can be produced by combining choose how much to consume and first-generation DSGE models. Current model development has proceeded along a number of labor and capital, can be viewed as un- invest as well as how much labor to lines, of which the most important are the addi- affected by changes in monetary policy supply. Importantly, the function that tion of more sophisticated financial markets and — independent of the level of interest specifies how consumption is valued more detailed depictions of labor markets using search theory. In terms of models employed at rates, the same amount of machines involves habit persistence, meaning various central banks, the model developed by that consumers value their current the Federal Reserve Bank of New York and one of the models used by the European Central level of consumption relative to previ- Bank include separate financial sectors. 4 This framework is based on Calvo. ous levels of consumption. This implies that consumers value a given level of consumption differently depending FIGURE on whether that level was less than or greater than the amount of consump- PRISM tion they experienced in the past. If that level of consumption corresponds to a relatively high amount, then the Households consumer is happier than if it corresponds to a relatively low amount. This aspect of behavior turns out to be a relatively important ingredient for the model’s ability to generate the type of Labor and economic persistence that is typically Capital Taxes found in U.S. economic data. Unlike the choice of consump- Output tion, which is fairly standard, the labor supply decision in PRISM is much Fiscal Policy different than is typically used in ba- Output Monetary Policy sic real business cycle models. These models view labor markets as purely Firms Government, Fed competitive, but in PRISM and most DSGE models, households are viewed as being able to influence wages in much the same fashion that firms set www.philadelphiafed.org Business Review Q2 2013 13 prices. They then supply all the labor As is true with most current and inflation. The shock reflects these demanded by firms at that wage. As is DSGE models, PRISM contains a deviations of actual policy from the the case with prices, only a subset of nonproductive government sector that Taylor rule. wages is adjusted in any period, and consumes resources, but that is gener- Model development is ongoing, the average wage is thus sticky. ally the extent to which fiscal policy is and although many models, including The evolution of the capital stock incorporated into the model. Monetary those being studied by staff at various is also determined by households’ in- policy is modeled as a simple Taylor Reserve Banks and the Board, share vestment decisions, and the accumula- rule in which interest rates respond to most of the above features, they do dif- tion of capital is subject to adjustment inflation relative to target, an output fer along many dimensions. Thus, the costs such as those that accompany the gap, and the past setting of the interest field of DSGE modeling provides a rich installation of new equipment. These rate. The output gap in PRISM is the set of models, which unsurprisingly costs are also random and affect the difference between current output and often present different interpretations efficiency of investment. The more the output that would occur in the of economic events. costs associated with adjusting the capital stock, the less new capital is obtained from any particular level of As is true with most current DSGE models, investment. This shock can be given PRISM contains a nonproductive government a financial interpretation (see the ar- sector that consumes resources, but that is ticle by Alejandro Justiniano, Giorgio Primiceri, and Andrea Tambalotti). In generally the extent to which fiscal policy is particular, when the financial system is incorporated into the model. not operating efficiently, it is more difficult for firms to purchase investment goods, and the allocation of invest- absence of any economic disturbances. USES OF THE MODELS ment also becomes less efficient. The That is, it is the difference between IN MONETARY POLICY authors show that a shock to the ef- current output and its trend. In this Once a DSGE model is estimated, ficiency with which firms transform in- regard, we find differences across vari- it can be used to provide economic vestment into increases in the stock of ous DSGE models, with some going forecasts and to identify the distur- capital is highly negatively correlated so far as to construct gaps based on bances that are driving the forecast. with the interest premiums charged to statistical procedures similar to those All central banks find it important to firms, and these premiums are related employed in actual statistical measures forecast economic activity when arriv- 6 to financial constraints. of the gap. The Taylor rule also speci- ing at a policy decision, and to that ex- Another common random dis- fies a gradual adjustment of policy to tent, these models provide another fore- turbance that influences households’ movements in inflation and the gap casting platform. Regarding the quality decisions involves shocks to the rate of and is also subject to a random distur- of the forecasts made with DSGE mod- time preference. This shock affects the bance to monetary policy. In reality, els, they are generally of similar quality degree to which households are willing the conduct of monetary policy is more to forecasts based on other types of to sacrifice current consumption and nuanced than the behavior specified forecasting methods or forecasts that 7 thereby increase saving, which then in the Taylor rule, with policymak- are more judgmental in nature. For allows the household to consume more ers reacting to more than just output example, a 2012 study by Marco Del in the future. As a result, shocks to the Negro and Frank Schorfheide indicates rate of time preference can be impor- that, at short horizons (one quarter), DSGE models do about as well as purely tant in generating differential growth 5 Although shocks to the wage markup are not patterns in consumption and invest- present in PRISM, most DSGE models feature statistical procedures when forecasting ment. Shocks to the value of leisure such shocks, which affect the costliness of labor. output and inflation, but at horizons of (which affect labor supply) are also 6 For a detailed discussion of various ways that one year, they do somewhat better. This featured in PRISM and most DSGE output gaps are measured, see the Business Review article by Roc Armenter and the study models. Shocks to leisure are intended by Michael Kiley. A particular DSGE model to capture any imperfections in labor that calculates a statistically based output gap is 7 However, forecasts that use various model markets beyond those involving wage the DSGE model being developed by staff at the restrictions in forming priors still generally out- Chicago Fed (see the article by Charles Evans perform those from DSGE models (see the 2004 rigidity.5 and coauthors). study by Del Negro and Schorfheide).

14 Q2 2013 Business Review www.philadelphiafed.org is also the message of the study by Maik can ask what the models predict if a model misspecification can lead to an Wolters, who additionally shows that disturbance was somewhat larger than incorrectly designed policy. taking forecast averages across various estimated or if it were to turn out to be Also, because none of the models DSGE models can improve their fore- more long-lived than usual. Doing so are literally true, they do not present a casting performance. lets policymakers gauge risks associated totally accurate depiction of the econo- The models can also be used to with particular economic events. my. However, looking at the output of benchmark policy, since one of their various models can help to clarify the forecasts is for the behavior of interest SOME WEAKNESSES extent of that misspecification. rates. Also, standard error bands can OF THE MODELS Of greater significance is the fact be placed around the forecasted path My overview would be incom- that some of the behavioral relation- of the interest rate, allowing policy- plete if I did not point out some of the ships in the models are not really makers to perceive the likelihood of inherent weaknesses of the current invariant to monetary policy. As men- a particular benchmark path. The generation of DSGE models. Perhaps tioned, the price-setting mechanism Riksbank employs its DSGE model for the most important is model mis- precludes changes in price-setting this purpose. specification. Currently, many of the behavior at different inflation rates. A relative strength of the DSGE restrictions imposed by the various Thus, policies that affect the behav- framework lies in its ability to iden- DSGE models are at odds with the ior of inflation are likely to affect the tify shocks. For example, many DSGE data. For example, the models specify actual economy in ways that the model models identify shocks associated with that, in the long run, variables such cannot capture. Thus, the implica- the impairment of financial markets as consumption, output, investment, tions drawn from the model may not as being primarily responsible for the and wages all grow at the same rate, be entirely accurate. This problem is most recent recession and the current which is somewhat at odds with the less severe if the variation in inflation slow recovery. Identifying the most data. One outgrowth of this type of associated with an alternative policy is important shocks in any given eco- misspecification is that many of the not very large, but the model’s predic- nomic episode is particularly important economic disturbances in the model tion will be less reliable if the variation for a monetary policymaker, since the must be very persistent in order to in inflation is significant. Thus, when optimal response to demand shocks is analyzing alternative policies, policy- often much different than the optimal A relative strength of makers should have more confidence response to supply shocks. Thus, it is in the model’s prediction when the important to identify what types of the DSGE framework alternative is closer to actual policy. economic disturbances are affecting lies in its ability to Furthermore, issues concerning the economy if a policy decision is to the identification of various parame- be a fully informed one. identify shocks. ters sometimes arise. By that I mean an DSGE models are also used to ex- occurrence when the data are not par- plore the effects of alternative policies. align the model with the data. Incor- ticularly informative about the value of Because all the sectors of the model rect estimation of the disturbances a parameter. In that case, the estimat- are formally linked together, along can affect the implications for how the ed value of the parameter will reflect with the assumption that the estimat- economy would react to a change in only the modeler’s prior belief about ed parameters are invariant to changes monetary policy. In a 2009 paper, Del the parameter no matter what that in policy, we can carry out policy ex- Negro and Schorfheide show that if prior belief happened to be. Hence, ercises that are easily interpreted.8 For the estimated DSGE model attributes very little is actually known about the example, we can analyze the effects of too much persistence to productiv- parameter. In cases like this, we need policies following alternative interest ity shocks, it implies that controlling to be particularly careful when assess- rate paths, paths that differ from the inflation would involve a monetary ing predictions of the model, especially model’s forecasted path. Further, we policy that responds overly aggres- if the parameter in question has an sively to departures of inflation from important effect on those predictions. target. That would not be the case if Finally, the models often lack im- 8 Formally, this means that the models are, the productivity disturbance was less portant sectors, such as a sophisticated in principle, not subject to Lucas’s famous persistent. Thus, when policymakers financial sector, and, as mentioned, critique regarding the inappropriateness of using the modeling of fiscal policy is quite relationships that are not based on a theoretical are deciding the best way to respond structural model to analyze policy changes. to departures of inflation from target, simplistic. These problems are not www.philadelphiafed.org Business Review Q2 2013 15 methodological, but they indicate that surrounding the outcomes of vari- Given the relative strengths and there is room for continuing evolution ous policy experiments. Thus, these weaknesses of current DSGE models, in this field of research. models are an important element of a they should be used in conjunction policymaker’s toolkit. They provide a with other forecasting methodologies SUMMARY coherent and internally consistent way and other models in combination with This article has outlined the basic of viewing the economy. other information and expertise that structure of a new class of models, The article has also pointed out policymakers bring to the table. In- DSGE models, which are currently be- some of the problems that currently deed, that is the way they are actually ing used to aid monetary policymakers exist within this class of models. It is being used by central banks around in many countries. They have proven important to understand that these the world.9 useful in forecasting, in identifying problems are not methodological, but key elements that are affecting the rather they reflect the current state of economy, and for conducting coun- the models. Development is ongoing, terfactual experiments that can help and many of the problems are currently 9 For an excellent and detailed discussion of how DSGE models are used in the context of policymakers understand both the being addressed in the next generation monetary policy at the Sveriges Riksbank, see likely outcomes and the uncertainty of models. the speech by Irma Rosenberg.

REFERENCES

Adolfson, Malin, Stefan Laseen, Jesper Del Negro, Marco, and Frank Schorfheide. Kiley, Michael. “Output Gaps,” Federal Linde, and Mattias Villani. “RAMSES “DSGE Model Based Forecasting,” Federal Reserve Board Finance and Economics — A New General Equilibrium Model Reserve Bank of New York Staff Report Discussion Series 2010-27 (2010). for Monetary Policy Analysis,” Sveriges 554 (March 2012). Riksbank Economic Review, 2 (2007), pp. Lucas, Robert. “Econometric Policy 5-40. Del Negro, Marco, and Frank Schorfheide. Evaluation: A Critique,” in K. Brunner “Monetary Policy Analysis with Potentially and A. Meltzer, eds., The Phillips Curve and Armenter, Roc. “Output Gaps: Uses and Misspecified Models,” American Economic Labor Markets. North-Holland, 1975. Limitations,” Federal Reserve Bank of Review, 99:4 (September 2009), pp. 1415- Philadelphia Business Review (First Quarter 50. Rosenberg, Irma. “The Monetary Policy 2011). Decision Process,” speech given at the Del Negro, Marco, and Frank Schorfheide. Riksbank, Stockholm, June 13, 2008. Bruback, Lief, and Tommy Sveen. “Nemo “Priors from General Equilibrium Models — A New Macro Model for Forecasting for VARs,” International Economic Review, Smets, Frank, Kai Christoffel, Guenter and Policy Analysis,” Norges Bank 45:2 (2004), pp. 643-73. Coenen, Roberto Motto, and Massimo Economic Bulletin, 80:1 (2009), pp. 39-47. Rostagna. “DSGE Models and Their Evans, Charles L., Jonas D.M. Fisher, Use at the ECB,” Journal of the Spanish Calvo, Guillarmo. “Staggered Contracts in Jeffrey R. Campbell, and Alejandro Economic Association (February 2010), pp. a Utility-Maximizing Framework,” Journal Justiniano. “Macroeconomic Effects of 51-65. of Monetary Economics, 12 (September Forward Guidance,” Brookings Papers on 1983), pp. 383-98. Economic Activity (2012). Wolters, Maik, “Evaluating Point and Density Forecasts of DSGE Models,” Chung, Hess T., Michael T. Kiley, and Federal Reserve Bank of Philadelphia. unpublished manuscript (March 2012). Jean-Pierre Laforte. “Documentation PRISM (DSGE Model); http://www. of the Dynamic Estimation-Based philadelphiafed.org/research-and-data/real- Optimization (EDO) Model of the time-center/PRISM/. U.S. Economy: 2010 Version,” Federal Reserve Board Finance and Economic Justiniano, Alejandro, Giorgio Primiceri, Discussion Series, 2010-29 (May 2010); and Andrea Tambalotti. “Investment http://www.federalreserve.gov/pubs/ Shocks and the Relative Price of feds/2010/201029/201029pap.pdf. Investment,” Review of Economic Dynamics, 14:1 (2011), pp. 102-21.

16 Q2 2013 Business Review www.philadelphiafed.org The Diverse Impacts of the Great Recession* BY Makoto Nakajima

he Great Recession had a large negative but it was still 16 percent lower than impact on the U.S. economy. Asset prices, income in 2007. A significant part of the decline in income was caused by a T most notably stock and house prices, declined rise in the unemployment rate. Figure substantially, resulting in a loss in wealth 1 shows how the unemployment rate and average income changed during for many American households. In this article, Makoto the Great Recession. The unemploy- Nakajima documents how diverse households were ment rate surged, from 4.7 percent in the fall of 2007 to 10 percent at its affected in a variety of dimensions during the Great peak in October 2009. Recession, in particular between 2007 and 2009, using Asset prices, most notably stock and house prices, declined substan- newly available data from the 2007-2009 Survey of tially during the Great Recession. This Consumer Finances. He discusses why it is important to decline in asset prices caused a loss in wealth for many American households. look at the data on households, rather than focusing on As for stock prices, Figure 2 shows that the aggregate data, and he reviews some recent studies the S&P 500 dropped from 1,496 in the last quarter of 2007 to 808 in the that look at the recession’s diverse effects on different first quarter of 2009, before recovering types of households. to around 1,400. The figure also shows how house prices declined. The average house price in 20 major metropoli- tan areas dropped by 34 percent from The Great Recession, which manuel Saez, average family income its peak in 2006 and has remained low began in December 2007, had a large (excluding capital gains) dropped by 17 since then.2 negative impact on the U.S. economy.1 percent between 2007 and 2009. Aver- In this article, I will document According to a recent study by Em- age income recovered slightly in 2011, how diverse households were affected in a variety of dimensions during the Great Recession, in particular between 1 In this article, I do not explain why stock pric- Krueger, and Jose-Victor Rios-Rull, argue that 2007 and 2009, using newly available es and house prices dropped significantly during shocks to economic productivity or demand the Great Recession. Some economists, includ- spilled over to the stock and housing markets. data from the 2007-2009 Survey of ing Andy Glover, Jonathan Heathcote, Dirk Nobuhiro Kiyotaki, Alexander Michaelides, Consumer Finances (SCF). The SCF and Kalin Nikolov analyze how such shocks provides detailed information on the to the economy become amplified and have a large impact on asset prices. Other hypotheses finances of U.S. households, and the exist. For example, Roger Farmer argues that Makoto special panel data allow us to compare changes in the beliefs of the market caused the Nakajima is a decline in housing and stock markets, which the same respondents between 2007 senior economist spilled over to the rest of the economy. Ulf von and 2009. While we might also like to in the Research Lilienfeld-Toal and Dilip Mookherjee argue that compare the fate of households over Department of the consumer bankruptcy law reform in 2005 the Philadelphia triggered the decline in house prices. the boom period before the Great Re- Fed. This article cession, the panel data from the SCF is available free of charge *The views expressed here are those of the at www. author and do not necessarily represent 2 The Case-Shiller Composite-20 index is used. philadelphiafed.org/research-and-data/ the views of the Federal Reserve Bank of The Case-Shiller national house price index fell publications/working-papers/. Philadelphia or the Federal Reserve System. similarly. www.philadelphiafed.org Business Review Q2 2013 17 are not available before 2007.3 data on households instead of focusing fall in average family income and the Why is it important to look at on the aggregate data? Although the decline in asset prices were large, behind the headline numbers, the effects

3 The regular Survey of Consumer Finances of the Great Recession varied greatly (SCF) has been conducted every three years surveys. However, families who participated in across households. One reason is that starting in 1983 to provide detailed information the 2007 survey were reinterviewed in 2009 in on the finances of U.S. families. Data from the order to capture how those households had been different households suffered differ- SCF are widely used in economic analyses. In financially affected by the Great Recession. ent degrees of income loss. Moreover, the most recent survey, about 6,500 families The paper by Jesse Bricker, Brian Bucks, Arthur different households were affected dif- were interviewed. Usually, the regular SCF does Kennickell, Traci Mach, and Kevin Moore sum- not follow the same households across different marizes the results of the 2007-2009 SCF. ferently by the decline in asset prices because households differed in the amount and composition of wealth FIGURE 1 when the Great Recession started. For Average Income and Unemployment Rate example, a household in Las Vegas (where the house price index has declined by 62 percent since 2006) U.S. dollars in 2010 Percent that owned a house and invested most 65,000 11.0 of its assets in stocks suffered a larger 10.0 loss in wealth than another house- 60,000 9.0 hold that was renting in Dallas (where 8.0 the house price index declined by 6 55,000 7.0 percent) and kept most of its assets in bank accounts. Differences in income 6.0 and wealth at the time of the Great 50,000 5.0 Recession are also tied, in part, to Unemployment rate (right axis) 4.0 households having different earnings Average income (left axis) 45,000 3.0 histories as well as different choices for 2007 2008 2009 2010 saving and investment. Year In response to the severe recession, economists have been trying Source: Saez (2012) and Bureau of Labor Statistics to better understand the recession’s diverse effects on different types of FIGURE 2 households, and I will review some recent studies. It is easy to understand Stock Market and House Price Indexes that households that suffered a larger loss of income or portfolio values suf- Index Index (2006 Q1=100) fered greatly from the recession. How- 1,600 110 S&P 500 (left axis) ever, Wenli Li and Rui Yao argue that Case-Shiller Home Price Index (right axis) when house prices decline, younger 1,400 100 renters benefit because they could buy houses at cheaper prices. On the 1,200 90 other hand, older homeowners, who tend to be sellers of houses, suffer from 1,000 80 a decline in house prices. As Glover and coauthors note, such an effect 800 70 was stronger during the Great Reces- sion because the prices of houses and 600 60 2006 2007 2008 2009 2010 2011 2012 financial assets fell significantly. They Year investigate how the welfare of different types of households has been affected differently by the Great Recession. On Source: Standard & Poor’s the other hand, Sewon Hur argues that

18 Q2 2013 Business Review www.philadelphiafed.org young households might not be able to after retirement. After retirement, shows the proportion of households in seize the opportunity to buy housing households use their savings to supple- each age group with a positive amount and other assets at depressed prices ment their (lower) income, gradually of housing assets, stocks, and business- because young households typically do reducing savings. es. The homeownership rate was 71 not have a lot of savings with which to The composition of wealth also percent in 2007 overall, but it was only buy assets, and it is difficult to borrow, shows a distinctive pattern over the life 28 percent for households in their 20s.5 especially during recessions. cycle. Let’s start with housing. Figure 5 The homeownership rate increases to

LIFE CYCLE AND WEALTH BEFORE THE GREAT FIGURE 3 RECESSION Before looking into how differ- Proportion of Households ent households have been affected by in Different Age Groups in 2007 the Great Recession, let’s look at how households differed on the eve of the Proportion 0.25 Great Recession. As you can see in

Figure 3, there were more households 0.20 whose heads were in their 40s and 50s in 2007 than in other age groups. 0.15 Net wealth (which is the sum of all assets, including the value of 0.10 houses, net of the sum of all debts) dif- fers over one’s life cycle. It is relatively 0.05 low for young households but keeps increasing during the working life of 0.00 20s 30s 40s 50s 60s 70+ households, up to around age 65, and Age declines after retirement. We can see such a pattern in Figure 4. Why Source: Survey of Consumer Finances, 2007-09 does the life cycle profile look like this? Franco Modigliani and Richard Brumberg provide a simple theory of FIGURE 4 the life cycle of a household.4 Young households, whose income is lim- Income, Housing, and Total Wealth ited, spend most of their income for of Households in 2007 consumption expenditures, leaving little savings to accumulate wealth. U.S. dollars in 2009 However, as households age and their 300,000 income increases, they start saving to Median income prepare for retirement. Figure 4 shows 250,000 Median wealth Median housing assets that both wealth and income go up 200,000 for households between their 20s and 50s. Saving for retirement is desirable 150,000 because after retirement, income is 100,000 typically lower than it is during middle age, when income is typically the high- 50,000 est over the life cycle. Households do 0 not want to have less money to spend 20s 30s 40s 50s 60s 70+ Age

4 Satyajit Chatterjee’s Business Review article provides a more detailed explanation of the Source: Survey of Consumer Finances, 2007-09 theory. www.philadelphiafed.org Business Review Q2 2013 19 67 percent for households in their 30s pared with their wealth holdings. In more households have accumulated and reaches 87 percent for those in other words, these young households enough wealth to make a down pay- their 60s, before shifting down to 83 are highly leveraged. ment. When these households pur- percent for those above 70. We can see Let’s go back to the comparison chase their first house, many of them the hump-shaped pattern in Figure 5. between Figure 5 and Figure 6. The have to invest most of their wealth in The proportion of wealth invested in homeownership rate picks up between home equity in the form of a down housing assets (shown in Figure 6) is the 20s and the 30s because, by then, payment. That’s why the proportion also hump shaped, but the peak comes much earlier than it does for wealth or the homeownership rate. Figure 6 FIGURE 5 shows the portfolio allocation grouped by different types of assets of house- Percentage of Households with Homes, Stocks, holds with median wealth.6 All values and Businesses of assets and debt are normalized by the wealth holdings of the median Percentage 100 households. For example, the value of Homeownership rate Stock ownership rate housing assets for median households 80 Business ownership rate in their 20s is 1.63, which implies that the value of housing assets of the me- 60 dian households is 163 percent of the 40 value of the wealth of these households. Debts are shown in negative 20 value. “Safe assets” include all assets except housing, stocks, and businesses, 0 20s 30s 40s 50s 60s 70+ e.g., checking and saving accounts, Age U.S. Treasury bills, and saving bonds. Therefore, for each age group, the sum Source: Survey of Consumer Finances, 2007-09 across all assets and debts is one. In other words, if the bar in Figure 6 is stretched long, it means the groups of FIGURE 6 households are taking a leveraged posi- tion, by borrowing and using the extra Portfolio Allocation by Median Households money to have more assets. (relative to total value of wealth) What can we see in Figure 6? Proportion relative to net wealth First, the proportion of wealth invested 4 in housing increases between the 20s 3 and the 30s and declines after that. Second, households in their 20s and 2 30s borrow significant amounts com- 1

5 The homeownership rate remained stable at 0 around 64 percent between 1965 and 1995, be- Safe asset fore rising to around 70 percent. However, the -1 hump-shaped pattern described in the article Debt remained stable. Matthew Chambers, Carlos Business -2 Stock Garriga, and Don E. Schlagenhauf investigate Housing reasons behind the increase. -3 6 Instead of looking at the single household with 20s 30s 40s 50s 60s 70+ median wealth, I take the average of households Age in the middle quintile (between 40 and 60 percent when ranked by wealth holdings). By doing this, I can avoid the situation that the results are affected by the behavior of one household. Source: Survey of Consumer Finances, 2007-09 See the next footnote as well.

20 Q2 2013 Business Review www.philadelphiafed.org of wealth invested in housing peaks The proportion of households that held almost zero wealth. The median for households in their 30s. However, have an equity interest in a privately wealth among the wealthiest 1 percent after households buy their first house, held business also exhibits a hump was almost $13 million.8 they repay the mortgage and start shape, as shown in Figure 5. Among Figures 7 and 8 show that there accumulating financial assets, which households in their 20s, only 6 percent is a substantial difference in stock decreases housing assets as a propor- have business equity, while the propor- holdings across households with dif- tion of household wealth. As we can tion is highest among households in ferent amounts of wealth. Among the see in Figure 6, for median households their 60s, at 17 percent. The propor- wealthiest 20 percent, 90 percent hold in their 20s to 40s, the average value of tion is 5 percent for households age stocks. On the other hand, among the housing assets is higher than the value 70 and above. I will come back to the households in the bottom 20 percent of their net wealth. As households wealth allocated to businesses in the in terms of wealth in 2007, only 17 continue to accumulate wealth for next section, since investment in busi- percent own stocks. Richer households retirement, the proportion of the value ness is closely related to large wealth tend to invest more in stocks as well. of housing assets included in house- holdings. Figure 6 shows that the pro- The median value of stocks held by the holds’ wealth keeps shrinking. In other portion of wealth invested in business- wealthiest 20 percent of households words, households keep deleveraging. es by median households is less than 5 is $183,000, while the median stock The proportion of households percent for all age groups. value is zero among the bottom 20 per- with a positive amount of stocks (in- cent and the median stock value of the cluding directly held stocks as well as RICH AND POOR ON THE EVE middle quintile of wealth distribution those held indirectly through mutual OF THE GREAT RECESSION is about $1,000. funds, retirement funds, etc.) is also There are large differences across hump shaped, as in Figure 5. The households if we look at them in dif- proportion is 37 percent for households ferent quintiles of wealth distribution.7 7 A quintile is one-fifth of all households. The in their 20s, peaks at 64 percent for As shown in Figure 7, the amount of first quintile represents the bottom 20 percent of households when households are sorted by households in their 50s, and then goes assets held by households in different the amount of wealth holdings. In other words, down to 44 percent for households quintiles of the wealth distribution the first quintile includes households with the least amount of wealth, and the fifth quintile in their 70s. Why is it hump shaped? differed significantly in 2007. The includes the top 20 percent of the wealthiest Annette Vissing-Jorgensen argues that median wealth holding of the wealthi- households. many households do not hold stocks est 20 percent was $972,000, while the 8 It was about $11 million in 2004, according to because of the costs of participating least wealthy 20 percent of households the SCF. in the stock market. Since younger households tend to have lower wealth, they tend to stay away from the stock FIGURE 7 market because the cost of participation is too high for the small gain that Asset Holdings for Different Wealth Quintiles households expect from investing in U.S. dollars in 2009 the stock market. Young households 1,000,000 also want to use their money to own Median wealth housing rather than to invest in stocks. 800,000 Median housing assets On the other hand, older households Median stock value withdraw from the stock market to 600,000 reduce their exposure to risky assets. In terms of the proportion of wealth 400,000 invested in stocks, the size is relatively small, as seen in Figure 6. Average 200,000 households invest relatively small proportions of their wealth in stocks. 0 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile For example, the proportion is about 15 percent for median households in their 30s to 50s and 10 percent among Source: Survey of Consumer Finances, 2007-09 households in their 60s. www.philadelphiafed.org Business Review Q2 2013 21 The homeownership rate is also higher for wealthier households FIGURE 8 (Figure 8). Among the top 20 percent in wealth holdings, 97 percent Proportion of Households with Home, Stocks, are homeowners. On the other hand, and Businesses in Different Wealth Quintiles the homeownership rate was 13 per- Proportion cent for the bottom 20 percent of the 1.0 wealth distribution in 2007. Naturally, Homeownership rate the median value of housing assets is 0.8 Rate with stocks Rate with business higher for wealthier households (Figure 7). It is $518,000 for the wealthiest 20 0.6 percent, while it is zero for the bottom 20 percent. However, the propor- 0.4 tion of wealth invested in housing is 0.2 decreasing as a share of household wealth among homeowners, precisely 0 because households with lower wealth 1st quintile 2nd quintile 3rd quintile 4th quintile 5th quintile have to spend more of their wealth on a house in order to buy one. For example, among households in the Source: Survey of Consumer Finances, 2007-09 middle quintile, the value of housing relative to wealth is 115 percent. On the other hand, the ratio is only 53 for older and wealthier households, ed. If capital gains are included, aver- percent among the top 20 percent of which tend to own larger houses. age income dropped by 17 percent.9 the wealth distribution. However, in relative terms, younger Andy Glover and coauthors computed The proportion of households that homeowners, who tend to invest a that overall average earnings declined own businesses increases significantly larger proportion of their wealth in by 8.3 percent, according to the Cur- with the level of wealth (Figure 8). In housing, suffer the most in terms of rent Population Survey (CPS). More- other words, wealthier households are the damage relative to their wealth. over, Glover and coauthors computed more likely to be entrepreneurs. For Remember Figure 6, which shows that that earnings of households in their example, 33 percent of the wealthiest younger households tend to be highly 20s declined by 11 percent, while earn- 20 percent of households own interests leveraged. As for other assets such as ings of households in their 60s dropped in business, while the ratio is less than stocks, again, middle-aged and older by only 6 percent. These facts are con- 2 percent among the least wealthy 20 households, especially the wealthy ones sistent with the ones presented by Mi- percent. The ratio is even higher for who invest more in the stock market, chael Elsby, Bart Hobijn, and Ayşegül the wealthiest 1 percent: 74 percent of suffer from a decline in stock prices. Şahin, who report that, in recessions, these households invest in businesses. The unfavorable business environment the unemployment rate rises more for The proportion of wealth invested in during recessions damages the wealthi- younger workers. businesses also increases significantly est households, which are more likely Elsby and coauthors also report with the level of wealth (Figure 7). to own businesses, the most. that the unemployment rate of workers with less education tends to rise WHAT SHOULD WE EXPECT? THE GREAT RECESSION’S more during recessions, including the From the way assets are distrib- DIVERSE EFFECTS ON INCOME Great Recession. This fact implies that uted, one can guess how changes in Before looking at wealth, let’s start workers with relatively lower levels of asset prices affect different households with income. The Great Recession education (and lower income) suffered differently. When house prices drop, had a large effect on income. Accord- a larger percentage drop in income middle-aged and older households, ing to a recent study by Saez that uses during the Great Recession. especially wealthy ones, suffer more data on individual tax returns, average because they are more likely to own a income per family in the U.S. declined 9 The drop was larger if capital gains are includ- house. In terms of the absolute level, by 11 percent between 2007 and 2009 ed because capital gains tend to react strongly the negative effect on wealth is larger if income from capital gains is exclud- to economic booms and recessions.

22 Q2 2013 Business Review www.philadelphiafed.org In contrast, Saez reports that, in track of the same households only in components of wealth individually. general, top income earners experience these two years, housing prices con- The average value of housing assets a larger percentage decline in income tinued to stagnate even after 2009. In dropped by 13 percent, from $262,000 from recessions. The explanation is How About 2010?, I compare house- to $228,000. The size of the drop is that the source of a large part of in- holds’ income and wealth in 2009 and smaller than the size of the drop in come for top income earners is capital 2010, using the newly available data the national house price index during gains. Saez computed that the top 1 from the SCF, although a direct com- the interval between the two surveys percent of the income distribution parison is difficult because the 2010 (19 percent). There are two reasons for suffered an income loss as large as 36 SCF does not keep track of the same this. First, the value of housing assets percent between 2007 and 2009, while households as in 2007 and 2009. is self-reported in the SCF, so there is the income loss was lower in propor- Housing. Let’s look at important possibly an upward bias, especially in a tion for the rest, at 12 percent. In sum, between 2007 and 2009, U.S. households suffered a large drop How About 2010? in income. The groups of households that suffered a larger loss than aver- he table compares the data on income and wealth across the age were younger households, lower- 2007, 2009, and 2010 Survey of Consumer Finances (SCF) pro- income households in each age group, vided by the Federal Reserve Board. Note that the households and extremely wealthy households. T included in computing the statistics are different across the Retired households, many of whom no 2007-2009 SCF and the 2010 SCF. The 2007-2009 SCF includes longer rely on labor income, suffered households that were age 20-99 in 2007 and surveyed in both 2007 and 2009. the least in terms of a percentage de- On the other hand, households between ages 20 and 99 in 2010 are included cline in income. in the 2010 SCF. Although housing prices and stock prices recovered somewhat between 2009 and 2010, median total net wealth dropped from $97,000 HOUSEHOLDS’ WEALTH IN to $76,000. Median housing assets declined slightly, from $180,000 in 2009 THE GREAT RECESSION to $176,000 in 2010. Median income declined as well, from $50,000 in 2009 In this section, I document how to $45,000 in 2010. The proportion of households that own housing and that wealth and its components changed own stocks also declined. However, a large part of these changes appears to be between 2007 and 2009, using the generated by differences in the households included in the SCFs. In particular, SCF. According to the SCF, the statistics in 2009 tend to be higher because 2009 data do not include households average net wealth of all households that were younger than 20 in 2007 or moved residence between 2007 and 2009 decreased from $595,000 in 2007 and thus were not followed in 2009. These households tend to be younger and to $481,000 in 2009, a 19 percent thus earn less and hold less wealth. As evidence, the Census Bureau reports that ($114,000) decline. Median wealth the homeownership rates in 2009 and 2010 were 67.4 percent and 66.9 percent, declined even more, from $126,000 in respectively. This homeownership rate is substantially lower than the homeown- 2007 to $97,000, a 23 percent decline ership rate in the SCF in 2009 (71.8 percent) but is closer to the homeownership ($29,000). For comparison, accord- rate in the 2010 SCF (68.9 percent). ing to the 2004 SCF, median house- Comparison Between 2007, 2009, and 2010 hold wealth was $107,000. Simply put, between 2007 and 2009, more than 2007 2009 2010 the gains in wealth between 2004 Median income (dollars) 50,000 50,000 45,000 and 2007 and about one-fifth of the Median total wealth (dollars) 126,000 97,000 76,000 wealth held by households in 2007 Median house value (dollars) 207,000 180,000 176,000 disappeared. For comparison, aver- Homeownership rate (%) 71.0 71.8 68.9 age household earnings (wage income) declined by 3 percent, from $56,000 to Proportion of stockholders (%) 53.7 55.6 50 $54,000, and average household total income dropped by 9 percent, from Note: Income and wealth are in 2009 dollars. For 2007 and 2009 data, households of age 20-99 in 2007 and surveyed in both 2007 and 2009 are included, while all households of $89,000 to $81,000. age 20-99 in 2010 are included in 2010 data. Although I compare the data from 2007 and 2009 because the SCF kept Source: Survey of Consumer Finances, 2007-2009 and 2010 www.philadelphiafed.org Business Review Q2 2013 23 down market. Households interviewed assets, which are defined as total as- represent the median wealth in 2007 for the survey might tend to think (or sets minus the value of housing assets, and 2009, respectively. Figure 9 also believe) that the value of their house is stocks, and businesses, was also rela- shows how the median value of hous- higher than it actually is. Second, the tively stable, at $175,000 in 2007 and ing assets and stocks changed between majority of households are at a stage $165,000 in 2009 (a 6 percent decline). 2007 and 2009. Figure 10 exhibits in life during which they are increas- This is not surprising, since the prices the changes in mean value of wealth, ing their holdings of housing assets. of safe assets such as bank accounts housing, and stocks. Note that we are talking about the value of the houses that households Households in their 30s and 40s suffered a own. If households buy a house for the first time or move up to a larger house, large loss in terms of median wealth between the value of the house owned by the 2007 and 2009. household probably increased, even if the same house was cheaper in 2009 and Treasury bills remained relatively Looking at different households compared with 2007. The median stable during the Great Recession.10 separately in Figures 9 and 10, we see value of housing assets declined less, Life Cycle. Figures 9 to 12 exhibit that average wealth declined for all age from $135,000 in 2007 to $125,000 in how households in different age groups groups between 2007 and 2009. How- 2009, a 7 percent decline. were affected during the Great Reces- ever, there are interesting differences Stocks. Between 2007 and 2009, sion. For example, in Figure 9, each across different age groups. the total value of stocks held directly grey line represents how the median First, the loss of wealth suffered or indirectly per household dropped by wealth of one age group changed by households headed by those in 29 percent, from $125,000 to $88,000. between 2007 and 2009; the points their 20s was limited in terms of the The total value of directly held stocks on the left and right side of each line absolute level. The loss, however, was (which do not include those held by large relative to the wealth they had pension funds or mutual funds) per in 2007. The mean value of wealth for 10 household dropped even more, by 37 In 2004, the total value of nonhousing assets per household was $368,000 (in 2009 U.S. dol- households in their 20s dropped by 23 percent, from $45,000 to $28,000. The lars). The per-household debt was $90,000. The percent between 2007 and 2009. The drop in the average value of stocks is value of safe assets per household was $159,000. median wealth held by households in Roughly speaking, the values in 2004 are not far consistent with the size of the drop from those in 2009. their 20s declined by 14 percent, which in the stock market index. Although these numbers are large, the long-run effects of this drop are probably lim- FIGURE 9 ited, because, as seen in Figure 2, the stock market rebounded strongly after Changes in Median Value of Wealth, Housing, 2009. As long as households were able and Stocks, 2007-09 to wait until the stock market recov- U.S. dollars in 2009 ered, they were able to minimize the 300,000 damage caused by the temporary slump Wealth in the stock market. The average value 250,000 Housing Stocks of businesses owned also dropped 200,000 sharply, by 23 percent, from $135,000 to $104,000. 150,000 Financial Assets and Debt. The 100,000 total value of nonhousing assets per household, which includes stocks, 50,000 business interests, and other financial assets, declined by 18 percent, from 0 20s 30s 40s 50s 60s 70+ $435,000 to $357,000. On the other Age hand, the average size of debt was stable: $103,000 in 2007 and $104,000 Source: Survey of Consumer Finances, 2007-09 in 2009. The average total value of safe

24 Q2 2013 Business Review www.philadelphiafed.org was smaller than the size of the decline ticular, since less than half of house- suffered a large loss in wealth relative in the median wealth of all households holds in their 20s own a home, the to their wealth holdings, because they (23 percent). This is mainly due to the household with median wealth was a were highly leveraged (taking out a characteristics of the median house- renter and did not suffer from a decline mortgage that is large relative to their hold among those in their 20s. In par- in house prices, while homeowners wealth holdings to buy a house). The mean value of housing assets increased slightly for this age group, but that is FIGURE 10 because they were at the stage in life during which they were buying houses. Changes in Mean Value of Wealth, Housing, Figure 11 shows that homeownership and Stocks, 2007-09 for households in their 20s increased even during the Great Recession and U.S. dollars in 2009 that more and more of them started 1,200,000 participating in the stock market. For Wealth households ages 20 to 29, the life-cycle 1,000,000 Housing effect strongly influences the changes Stocks 800,000 in the data. Figure 12 is the counterpart to 600,000 Figure 6. Figure 12, which shows how the proportion of the value of housing 400,000 assets relative to net wealth for median 200,000 households changed between 2007 and 2009, is consistent with the fact 0 that younger households were buying 20s 30s 40s 50s 60s 70+ Age houses even during the Great Reces- sion. We can see that the proportion of wealth invested in housing increased Source: Survey of Consumer Finances, 2007-09 between 2007 and 2009 for households in their 20s. The value of debt relative to wealth also increased. Although we FIGURE 11 often hear that American households have been deleveraging (reducing debt) Changes in Ratio of Households with since the onset of the Great Recession, young households were still leveraging, Housing, Stocks, and Businesses, 2007-09 implying that, for these households,

Proportion the life-cycle effect (borrowing and 1.0 buying houses when they are young) Housing has dominated the deleveraging in 0.8 Stocks Business which older households were engaged. Figure 12 also shows that the propor- 0.6 tion of wealth invested in stocks or businesses by households with median 0.4 wealth remained low during 2007- 2009. 0.2 Households in their 30s and 40s

0 suffered a large loss in terms of median 20s 30s 40s 50s 60s 70+ wealth between 2007 and 2009 (Figure Age 9). Median wealth declined by 35 percent and 29 percent for households in Source: Survey of Consumer Finances, 2007-09 their 30s and 40s, respectively. Their wealth declined even though stock www.philadelphiafed.org Business Review Q2 2013 25 value of their assets fell, and they were FIGURE 12 actively reducing wealth. Figure 12 is consistent with such an interpretation; Changes in Portfolio Allocation by the proportion of wealth allocated Households with Median Wealth to housing by older households with

Proportion relative to net wealth median wealth declined between 2007 3 and 2009, albeit slightly. Turning to stock holdings, Fig- 2 ure 10 shows that the proportion of 1 households older than 70 with stocks 0 declined between 2007 and 2009. In

-1 other words, households age 70 and above were selling stocks. Therefore, -2 Housing Stocks the declining value of stocks among -3 Business older households exaggerates the loss Debt -4 suffered by these households because 20s 30s 40s 50s 60s 70+ they were actively selling stocks. Age However, as shown in Figure 12, households with median wealth do not Survey of Consumer Finances, 2007-09 invest much in stocks. On the other hand, the homeownership rate did not drop during 2007-2009, implying that holdings are limited, especially for of households in their 50s and 60s homeowners age 70 and above suffered households in their 30s. Their wealth declined by 20 percent and 13 percent, a loss in the value of their housing.11 declined significantly mainly because respectively. On the other hand, their Wealth Distribution. There is the median household is a homeowner mean wealth declined more than their also a large diversity in how different with a highly leveraged portfolio and median wealth, but this was due to households in different parts of the thus exposed to a higher risk of declin- a large decline in the value of busi- wealth distribution were affected by ing housing and stock prices. On the nesses, which was concentrated among the Great Recession. Figure 13 shows other hand, mean wealth declined less a small number of households in their the percentage changes in the mean because changes in wealth holdings 50s and 60s. Figure 11 shows that the value of wealth, housing, and stocks for of renters, who were not affected by proportion of households with busi- groups of households in different parts declining housing prices, affect the ness interests was highest within these of the wealth distribution. We can see mean more than the median. Figure age groups. Since the homeownership clearly that changes in wealth were sig- 12 shows that the proportion of wealth rate stabilized at ages 50s to 60s, the nificantly different for households with invested in housing by households with proportion of wealth allocated to hous- different levels of wealth. In particular, median wealth continued to increase ing assets remained relatively stable households with the lowest amount of because of the life-cycle effect. The between 2007 and 2009 (Figure 12). wealth increased their wealth hold- size of debt relative to wealth also The median wealth of house- ings between 2007 and 2009, mainly increased between 2007 and 2009 for holds aged 70 and above declined by because of life-cycle effects. These middle-aged households. 22 percent, which was larger than the households were in the life-cycle stage For households in their 50s and decline for households in their 50s during which they accumulate wealth. 60s, wealth declined between 2007 and and 60s. Mean wealth declined by 25 Between 2007 and 2009, the average 2009, as seen in Figures 9 and 10, but percent. An important part of this wealth held by the bottom 40 percent the loss was relatively small, because large decline was due to life-cycle pat- of the wealth distribution increased by these households had already accumu- terns; households age 70 and above 54 percent, from $13,000 to $20,000. lated wealth and invested a larger part were spending down their accumulated of their wealth in safer assets (see also wealth to support consumption expen- Figure 6). Therefore, their exposure ditures in retirement. In other words, 11 In an earlier Business Review article, I docu- mented how retirees decumulate wealth, with a to risky assets such as housing and the size of the decline for households focus on the distinction between housing and stocks was lower. The median wealth in retirement looks larger because the financial assets.

26 Q2 2013 Business Review www.philadelphiafed.org This is due in large part to an increase that the choice of the timing of the trivial channels that create winners in average holdings of housing as- comparison matters significantly. Even and losers. I will slice households along sets, which increased from $39,000 if a household lost during the Great various dimensions and discuss who to $44,000. The homeownership rate Recession, because the value of the as- gained and who lost from the Great among these households also in- sets that the household owns declined Recession; in particular, I will look creased, from 35 percent to 40 percent. between 2007 and 2009, the household at the large drops in income, house They increased their stock holdings, might have gained if the value of the prices, and stock prices. but stocks’ contribution to the increase assets in 2009 is compared with the Income. On average, households in wealth is limited because these value in, say, 2002. On the other hand, lost income from the Great Recession. households invested little in the stock households that purchased their house However, young and less educated market from the beginning. On the at the peak of house prices (around households tended to suffer a larger other hand, the average wealth among 2006-2007) lost value in their house percentage drop in income. Moreover, the top 20 percent in the wealth dis- without benefitting from the boom using Canadian data, Philip Oreopou- tribution declined by 21 percent, from that preceded the decline.12 The analy- los, Till von Wachter, and Andrew $2.5 million to $2.0 million. They sis here is limited in the sense that it Heisz show that college students gradu- experienced a loss in holdings of hous- cannot account for changes that hap- ating and entering the job market in a ing assets, stocks, and businesses. The pened before the Great Recession. recession suffer a large initial income loss was even more pronounced for As I have shown, households with loss, and the loss is persistent, lasting the wealthiest 1 percent; their wealth different levels of income or wealth as long as 10 years. On the other hand, dropped by 29 percent during 2007- and at a different stage of life were af- Saez shows that households at the top 2009, although about two-thirds of fected differently by the Great Reces- of the income distribution experienced them remained among the wealthiest sion. Moreover, there are some non- a larger percentage loss from the Great 1 percent even after the loss. Recession. Middle-aged households suffered less because more of them 12 WHO GAINED AND WHO LOST Many people, including economist Robert Shiller (who helped to develop the Case-Shiller have stable, full-time jobs. Not surpris- IN THE GREAT RECESSION house price index), perceive the substantial rise ingly, retired households suffered little. Who benefited and who suffered in housing prices before the Great Recession to Housing. Homeowners, especially be a “bubble.” Please see my previous Business from the Great Recession? Before Review article for a discussion of the “bubble” those who wanted to sell their house, going into details, let me emphasize theory of house prices. suffered from the drop in house prices. Younger homeowners suffered less FIGURE 13 because they could likely wait until house prices recover (if they ever do) Percentage Changes in Mean Value of Wealth, to sell. Although house prices have Housing, and Stocks, 2007-09 hit bottom and are finally rising, they remain a long way from their levels before the housing crash. Whether and Percentage changes between 2007 and 2009 120 how much homeowners suffer depends 100 Mean wealth on how fast and how much house Mean housing prices recover in the future. On the 80 Mean stock value 60 other hand, renters who were about to buy their first home or homeowners 40 planning to move up to a larger house 20 could buy houses at lower prices than 0 before the recession. Relatively young -20 renters and younger homeowners were -40 in this category. In their study, Wenli -60 1st and 2nd quintiles 3rd quintile 4th quintile 5th quintile Li and Rui Yao investigate these asym- metric effects of house-price changes. At the same time, they are likely to Source: Survey of Consumer Finances, 2007-09 have suffered a loss in income, and lost savings to be used for a down payment www.philadelphiafed.org Business Review Q2 2013 27 with the declining asset prices. There- ferent channel. How? If such a house- workers might have experienced a fore, whether these households gained, hold experienced a loss of income, decline in income during the Great all things considered, is not certain. even if it wanted to borrow money Recession, but they have more time Note that the timing of a home to avoid a large drop in consumption to bounce back, with possible booms purchase also matters. Homeown- expenditures, it might not be able to in the future canceling the Great Re- ers who purchased their house when do so if borrowing was difficult. A cession’s negative effects on income. house prices were still low might not household that wants to borrow but Older workers, on the other hand, have suffered too much from the Great cannot is called borrowing constrained. have a shorter time horizon because Recession, even with a large decline Households that are planning to buy they will retire sooner. in house prices, because the purchase houses or other assets suffer from the Welfare. All things above con- price was also low. To give an extreme borrowing constraint as well because, sidered, how did the Great Reces- example, the average price of new even though they want to buy houses sion affect the welfare (well-being) of homes was $229,000 in 2002, $314,000 or assets at depressed prices, they diverse households? The discussion in 2007, and $273,000 in 2010.13 For cannot do so because they have little above indicates that young households a person who purchased his house in wealth and are unable to borrow. This suffered more in terms of income, but 2002, selling in 2010 is worse than sell- point is emphasized in a recent paper older households suffered more from ing in 2007, but still his selling price by Sewon Hur. declining asset prices. Using a sophis- would be higher than the purchase Young and Old. Glover and co- ticated economic model, Glover and price. For a person who purchased his authors argue that age is an important coauthors computed that the size of house in 2007, that’s not the case.14 determinant of the impact of the Great the decline in the average welfare of On the other hand, homeowners who Recession, especially if the decline in households age 70 and above associ- purchased their house recently are the ones who suffered the most from the large decline in prices. Although house prices have hit bottom and Stocks. Similarly, households that are finally rising, they remain a long way from were about to sell stocks and could not their levels before the housing crash. Whether wait until the stock market recovered suffered from the Great Reces- and how much homeowners suffer depends sion. Older households, which tend on how fast and how much house prices to sell stocks to support consumption expenditures in retirement, were in recover in the future. this group. Relatively younger households and those experiencing income stock and house prices is temporary ated with the Great Recession was growth gained in this regard because and prices recover in the not-too-dis- equivalent to an 8 percent drop in they tend to be buyers of financial as- tant future. Under such circumstances, consumption every year for the rest sets, and they were able to buy assets at young households that have assets such of their lives. On the other hand, the depressed prices. as housing and stocks can hold on to decline in the welfare of young (20s) Wealth and Debt. Households these assets until prices recover and households was equivalent to a less with little wealth or those that were avoid losing wealth from the decline in than 0.5 percent decline in consump- heavily indebted did not suffer from asset prices. On the other hand, older tion every year. declining asset prices but could suffer households might not have time to Why did young households suf- from the Great Recession from a dif- wait until asset markets recover. Time fer less than older households? First, is especially important when they want young households are expected to live to sell the assets to support current longer. As long as the economy recov- consumption expenditures; holding on ers from the Great Recession in the 13 According to the Census Bureau, the median house price was $188,000 in 2002, $248,000 in to assets with depressed prices hurts future, the young can smooth out the 2007, and $222,000 in 2010. them because they might not be able losses from the Great Recession over to buy what they want if they do not 14 Here I assume that these are the prices with their lifetime. Second, young house- which households buy or sell their houses. Of sell these assets. holds tend to be accumulating assets, course, the first person “suffered” as well if he A similar argument can be made and thus they benefit from lower as- thought the value of his house was actually $314,000, but that’s a different story. about income. Relatively younger set prices. As we have seen, younger

28 Q2 2013 Business Review www.philadelphiafed.org households suffered a larger percentage Recession on different groups of house- more of their wealth in risky assets, loss in income on average, but accord- holds. In terms of income, young, low- although the majority of those wealthy ing to the calculation by Glover and er-income, and extremely high-income households remain relatively wealthy coauthors, this effect is weaker than households suffered a larger percentage even after experiencing a large loss. the two favorable effects for the young. decline. Moreover, a large decline in There are also nontrivial chan- However, we should remember that asset prices caused a larger drop in the nels. Older households tend to suffer this calculation did not take into ac- value of wealth for homeowners, stock- more because they tend to have less count the possibility that young house- holders, and business owners. In terms time to wait for asset prices to recover. holds with little or zero wealth could of age, middle-aged households tended On the other hand, young households suffer due to the borrowing constraint, to suffer a larger decline in wealth that buy assets indirectly benefit from as discussed above. because they tend to own those risky lower asset prices, but how much they assets more than younger and retired benefit from the Great Recession CONCLUSION households. The wealthiest households depends on whether they can actually In this article, I summarized the suffered more than the less wealthy in afford to buy these assets even after diverse economic impact of the Great proportion because they tend to invest suffering a loss in income.

REFERENCES

Bricker, Jesse, Brian Bucks, Arthur Glover, Andy, Jonathan Heathcote, Nakajima, Makoto. “Everything You Kennickell, Traci Mach, and Kevin Moore. Dirk Krueger, and Jose-Victor Rios-Rull. Always Wanted to Know about Reverse “Surveying the Aftermath of the Storm: “Intergenerational Redistribution in the Mortgages but Were Afraid to Ask,” Fed- Changes in Family Finances from 2007 Great Recession,” NBER Working Paper eral Reserve Bank of Philadelphia Business to 2009,” Federal Reserve Board, Finance 16924 (April 2011). Review (First Quarter 2012), pp. 19-31. and Economics Discussion Series 2011-17 (March 2011). Hur, Sewon. “The Lost Generation of the Oreopoulos, Philip, Till von Wachter, and Great Recession,” University of Minnesota Andrew Heisz. “The Short- and Long-Term Chambers, Matthew, Carlos Garriga, and Working Paper (February 2012). Career Effects of Graduating in a Reces- Don E. Schlagenhauf. “Accounting for sion,” American Economic Journal: Applied Changes in the Homeownership Rate,” Kiyotaki, Nobuhiro, Alexander Economics 4:1 (January 2012), pp. 1-29. International Economic Review, 50 Michaelides, and Kalin Nikolov. “Winners (August 2009), pp. 677-726. and Losers in Housing Markets,” Journal Saez, Emmanuel. “Striking It Richer: The of Money, Credit and Banking, 43 (March- Evolution of Top Incomes in the United Chatterjee, Satyajit. “The Peopling of April 2011), pp. 255-296. States (Updated with 2009 and 2010 Esti- Macroeconomics: Microeconomics of mates),” University of California, Berkeley Aggregate Consumer Expenditures,” Li, Wenli, and Rui Yao. “The Life-Cycle (March 2012). Federal Reserve Bank of Philadelphia Effects of House Price Changes,” Journal of Business Review (First Quarter 2009), Money, Credit and Banking, 36 (September Shiller, Robert J. Irrational Exuberance, 2nd pp. 1-10. 2007), pp. 1375-1409. edition. Princeton: Princeton University Press, 2005. Elsby, Michael W., Bart Hobijn, and Modigliani, Franco, and Richard H. Ayşegül Şahin. “The Labor Market in the Brumberg. “Utility Analysis and the Con- Vissing-Jorgensen, Annette. “Towards an Great Recession,” Brookings Papers on sumption Function: An Interpretation of Explanation of Household Portfolio Choice Economic Activity (Spring 2010), pp. 1-48. Cross-Section Data,” in Kenneth K. Kuri- Heterogeneity: Nonfinancial Income and hara, ed., Post-Keynesian Economics. New Participation Cost Structures,” NBER Farmer, Roger E. A. “The Stock Market Brunswick, NJ: Rutgers University Press, Working Paper 8884 (April 2002). Crash of 2008 Caused the Great Recession: 1954, pp. 388-436. Theory and Evidence,” Journal of Economic von Lilienfeld-Toal, Ulf, and Dilip Dynamics and Control, 36 (2012), Nakajima, Makoto. “Understanding Mookherjee. “How Did the U.S. Hous- pp. 693-707. House-Price Dynamics,” Federal Reserve ing Slump Begin? The Role of the 2005 Bank of Philadelphia Business Review (Sec- Bankruptcy Reform,” Boston University ond Quarter 2011), pp. 20-28. Working Paper (January 2011). www.philadelphiafed.org Business Review Q2 2013 29 Abstracts of research papers produced by the esearch ap economists at R R the Philadelphia Fed

You can find more Research Rap abstracts on our website at: www.philadelphiafed.org/research-and-data/ publications/research-rap/. Or view our working papers at: www.philadelphiafed.org/research-and-data/ publications/.

USING AN ASKING PRICE in agency CMOs using a traditional and a MECHANISM novel approach, and offers valuable lessons In many markets, sellers advertise learned when interpreting these risk mea- their goods with an asking price. This is a sures. Among these lessons is that to fully price at which the seller is willing to take understand the risks in agency CMOs, a full his goods off the market and trade im- bond-by-bond analysis is necessary and that mediately, although it is understood that a interest rate risk is not the only risk that buyer can submit an offer below the asking needs to be considered when conducting risk price and that this offer may be accepted if management with CMOs. the seller receives no better offers. Despite Working Paper 13-8, “Understanding and their prevalence in a variety of real-world Measuring Risks in Agency CMOs,” Nicholas markets, asking prices have received little Arcidiacono, Federal Reserve Bank of Philadel- attention in the academic literature. The phia; Larry Cordell, Federal Reserve Bank of authors construct an environment with a Philadelphia; Andrew Davidson, Andrew Da- few simple, realistic ingredients and demon- vidson & Company; and Alex Levin, Andrew strate that using an asking price is optimal: Davidson & Company It is the pricing mechanism that maximizes sellers’ revenues, and it implements the ef- INVESTIGATING WORKER FLOWS ficient outcome in equilibrium. The authors AND JOB FLOWS provide a complete characterization of this This paper studies the quantitative equilibrium and use it to explore the posi- properties of a multiple-worker firm match- tive implications of this pricing mechanism ing model with on-the-job search in which for transaction prices and allocations. heterogeneous firms operate decreasing- Working Paper 13-7, “Competing with returns-to-scale production technology. Asking Prices,” Benjamin Lester, Federal The authors focus on the model’s ability to Reserve Bank of Philadelphia; Ludo Visschers, replicate the business cycle features of job Universidad Carlos III; and Ronald Wolthoff, flows, worker flows between employment and University of Toronto unemployment, and job-to-job transitions. The calibrated model successfully replicates STUDYING THE AGENCY (1) countercyclical worker flows between em- CMO MARKET ployment and unemployment, (2) procyclical The agency CMO market, an often job-to-job transitions, and (3) opposite move- overlooked corner of mortgage finance, ments of job creation and destruction rates has experienced tremendous growth over over the business cycle. The cyclical proper- the past decade. This paper explains ties of worker flows between employment and the rationale behind the construction of unemployment differ from those of job flows, agency CMOs, quantifies risks embedded partly because of the presence of job-to-job

30 Q2 2013 Business Review www.philadelphiafed.org transitions. The authors also show, however, that job calibration and including all these features, the model flows measured by net employment changes differ can account for a sizable fraction of the elasticity of significantly from total worker separation and accession the value-output ratio to the first two moments of TFP rates because separations also occur at firms with posi- growth. tive net employment changes, and similarly, firms that Working Paper 13-10, “Risk, Economic Growth, and are shrinking on net may hire workers to partially offset the Value of U.S. Corporations,” by Luigi Bocola, Univer- attritions. The presence of job-to-job transitions is the sity of Pennsylvania, and Nils Gornemann, University of key to producing these differences. Pennsylvania Working Paper 13-9, “Worker Flows and Job Flows: A Quantitative Investigation,” Shigeru Fujita, Federal Reserve EXPANDING EMPLOYMENT THROUGH Bank of Philadelphia, and Makoto Nakajima, Federal STATE DEFICIT POLICIES Reserve Bank of Philadelphia Using a sample of the 48 mainland U.S. states for the period 1973-2009, we study the ability of U.S. THE LINK BETWEEN TFP GROWTH AND states to expand their own state employment through THE VALUE OF U.S. CORPORATIONS the use of state deficit policies. The analysis allows for This paper documents a strong association between the facts that U.S. states are part of a wider monetary total factor productivity (TFP) growth and the value and economic union with free factor mobility across all of U.S. corporations (measured as the value of equities states and that state residents and firms may purchase and net debt for the U.S. corporate sector) throughout goods from “neighboring” states. Those purchases may the postwar period. Persistent fluctuations in the first generate economic spillovers across neighbors. Esti- two moments of TFP growth predict two-thirds of the mates suggest that states can increase their own state medium-term variation in the value of U.S. corpora- employment by increasing their own deficits. There is tions relative to gross domestic product (henceforth evidence of spillovers to employment in neighboring value-output ratio). An increase in the conditional states defined by common cyclical patterns among state mean of TFP growth by 1 percent is associated with a economies. For large states, aggregate spillovers to their 21 percent increase in the value-output ratio, while this economic neighbors are approximately two-thirds of the indicator declines by 12 percent following a 1 percent large state’s job growth. Because of significant spillovers increase in the standard deviation of TFP growth. A and possible incentives to free ride, there is a potential possible explanation for these findings is that move- case to actively coordinate (i.e., centralize) the man- ments in the first two moments of aggregate productiv- agement of stabilization policies. Finally, when these ity affect the expectations that investors have regarding deficits are scheduled for repayment, the job effects of a future corporate payouts as well as their perceived risk. temporary increase in a state’s own deficits persist for at The authors develop a dynamic stochastic general equi- most one to two years, and there is evidence of a nega- librium model with the aim of verifying how sensible tive impact of state jobs. this interpretation is. The model features recursive pref- Working Paper 13-11, “Local Deficits and Local Jobs: erences for the households, Markov-Switching regimes Can U.S. States Stabilize Their Own Economies?,” Gerald in the first two moments of TFP growth, incomplete Carlino, Federal Reserve Bank of Philadelphia, and Robert information, and monopolistic rents. Under a plausible Inman, The Wharton School, University of Pennsylvania