Quantitative Easing in the 1930s

October 2018

Christopher Hanes Department of Economics State University of New York at Binghamton P.O. Box 6000 Binghamton, NY 13902 (607) 777-2572 [email protected]

Abstract: During the 1934-39 recovery from the U.S. Great Depression, overnight interest rates were usually at a lower bound while American monetary authorities followed policies related to today's debates on quantitative easing (QE): they tried to stabilize Treasury yields with open market operations; they created rapid growth in high-powered money; and they allowed transitory factors to affect high-powered money. I find relationships between these policies and bond yields that reveal a portfolio effect of short- duration asset supply on term premiums, and help explain why trend high-powered money growth was associated with recovery of real activity over 1934-39.

JEL codes: E43, E52, G12, N12, N22

Acknowledgements. For comments and suggestions, thanks to William English, John Fernald, James D. Hamilton, Barry Jones, Edward Nelson, Gary Richardson, Eric Swanson, Susan Wolcott and Wei Xiao; to participants in the Bank of San Francisco conference “The Past and Future of Monetary Policy,” March 2013; the Yale Economic History workshop; NBER DAE Summer 2014; of Cleveland workshop; Rutgers University Economic History Workshop. During the post-2009 recovery from the Great Recession, the Federal Reserve tried "quantitative easing" (QE) to influence bond yields while short-term rates appeared to be near a lower bound. In QE a acquires long-term bonds, usually Treasury bonds, in exchange for newly-created reserve balances. One immediate effect is to shorten the average duration of bonds left in the hands of the public.

Another is to increase the supply of high-powered (base) money at a time when the usual channel from money-supply changes to financial conditions, through changes in overnight rates, is ineffective.

Policymakers' goal was to reduce yields on bonds issued by private borrowers, such as corporate bonds, to spur spending and real activity. QE might lower yields through expectations of future overnight rates, by convincing market participants that policymakers will hold overnight rates low for longer. This is called the

“signalling channel.” But many advocates of QE claim QE squeezes down premiums that keep long-term rates above expected future overnight rates, through "portfolio-balance" effects.

Portfolio-balance effects are controversial (e.g. Greenlaw et. al. 2018). Event studies find statistically significant changes in yields around announcements of coming QE operations, but it is hard to rule out the possibility these were due to the signalling channel (Bernanke, Reinhart and Sack, 2004;

Krishnamurthy and Vissing-Jorgensen, 2011 and discussion; Gagnon et. al. 2011; Wright, 2012; Woodford

2013; Bauer and Rudebusch 2014). To disentangle portfolio effects from signalling, studies estimate time- series models of overnight-rate determination and factors other than QE that could affect term premiums

(Gagnon et. al. 2011; Hamilton and Wu 2012; D'Amico et. al. 2012). But their conclusions depend on debatable assumptions about the process determining the (Woodford 2013; Bauer and

Rudebusch, 2014). As a matter of theory, portfolio effects do not exist in standard representative-agent asset-pricing models (Eggertsson and Woodford 2003). In current literature many arguments for portfolio effects (e.g. Krishnamurthy and Vissing-Jorgensen, 2011; Gagnon et. al. 2011; D'Amico et. al. 2012) refer instead to the "preferred habitat" theory of Modigliani and Sutch (1966) as formalized in a model by

Vayanos and Vila (2009). In that model, an operation like QE can reduce term premiums because it reduces the average duration of bonds held by risk-averse investors, and hence reduces the extra return they require to bear duration risk. During the post-1933 recovery from the Great Depression, American short-term rates also appeared to be near a lower bound. American monetary authorities did not buy bonds to drive down long-term interest rates. But they did follow policies that increased high-powered money. And they often tried to influence bond yields by buying and selling Treasury bonds in the open market. Their goal was to stabilize

Treasury yields - to stop sudden increases or decreases in bond prices. In this paper I examine effects of those policies, guided by current views of portfolio effects. I help solve an old puzzle about the 1934-39 era. I also find relatively unambiguous evidence for portfolio effects and for central banks' ability to influence long-term yields, including corporate yields, through QE.

To apply current views of portfolio effects to 1934-39 I must account for the extremity of financial- market conditions in that era. After 2009 overnight lending markets remained active (though at diminished volume) with positive (though low) interest rates that varied from day to day (Gagnon and Sack 2014).

Through most of 1934-39, on the other hand, the return to safe overnight lending was simply zero. Under this condition, I show, a model following Vayanos and Vila (2009) implies that term premiums can be lowered by an increase in high-powered money, whether or not the increase is accompanied by a change in bond supply. That is because the public’s willingness to hold money - a safe asset paying no interest - is governed by a tradeoff of its relatively low return against its freedom from duration risk. Thus demand for such money is negatively related to the expected return to holding bonds; an increase in the outside supply of money lowers term premiums. For 1934-39, this implies that an increase (decrease) in high-powered money could cause a decrease (increase) in medium- and long-term bond yields.

Such a mechanism would help solve an old puzzle about 1934-39: trend growth in high-powered money appears to have affected real activity even though it had little if any effect on short-term nominal interest rates. Due to policies on international gold inflows and sterilization, trend high-powered money growth was rapid over 1934 through December 1936, about zero from December 1936 to late 1937, and rapid again after that. Trend growth in real activity followed the same trajectory after the usual lag of about six months in real effects of monetary-policy innovations (e.g. Christiano, Eichenbaum and Evans 2005):

3 output and employment grew rapidly after 1933 until the downturn of a recession in June 1937; real growth resumed in July 1938. To explain the coincidence it has been argued that high-powered money growth, or policies resulting in high-powered money growth, boosted real activity by creating expectations of future which lowered long-term real interest rates (e.g. Bernanke, Reinhart and Sack 2004:18-19;

Eggertsson and Pugsley 2006; Eggertsson 2008). According to Romer (1992: 775-76) "nominal interest rates were already so low that there was little scope for a monetary expansion to lower nominal rates further. Therefore, the main way that the monetary expansion could stimulate the economy was by generating expectations of inflation and thus causing a reduction in real interest rates." I show trend high- powered money growth was in fact accompanied by substantial declines in medium- and long-term nominal yields consistent with a portfolio effect of high-powered money growth on term premiums. Thus, whether or not trend high-powered money growth boosted expected inflation, it could have stimulated real activity simply by reducing longer-term nominal interest rates.

But was this mechanism actually present in the 1930s? To test that, I examine effects of factors causing short-term fluctuations in high-powered money. Throughout 1934-39, payment flows between the banking system and the Treasury’s Federal Reserve accounts (like the 1990s Treasury-payment shocks studied by Hamilton [1997]) created shocks to high-powered money that were essentially transitory and exogenous to financial-market conditions. Through portfolio effects, these shocks should have tended to affect bond yields. At the same time, also through portfolio effects, monetary authorities’ yield-stabilizing operations should have tended to counteract effects of these high-powered money shocks on yields. Yield- stabilizing operations were rare prior to July 1936. They became routine after that. Thus, if my hypothesis is correct, high-powered money shocks due to Treasury payments should be observed to affect bond yields from 1934 up to July 1936. Their effects should be weaker or absent after that. I examine weekly data on yields and high-powered money-supply factors and find this was indeed the case, for both Treasury and corporate yields. This is relatively unambiguous evidence for portfolio effects because the apparent relationship between yields and high-powered money shocks up to July 1936 cannot be accounted for by

4 the signalling channel. Within 1934-36, high-powered money shocks due to Treasury payments signalled nothing about future monetary policy or overnight rates. Finally, I check whether their apparent relationship to yields over 1934-36 was due to portfolio effects of correlated changes in bond supply, rather than shocks to high-powered money. To do this I use newly constructed weekly series on Treasury debt issuance. I find distinct evidence of a high-powered money effect.

To begin, I review current literature on portfolio effects of QE. Then I use a version of the Vayanos and Vila (2009) model to illustrate the portfolio effect of high-powered money on bond yields that exists when there is no return to safe overnight lending. In the third section I review the history of monetary policy, short-term interest rates and bond yields in the 1930s, and explain how a high-powered money portfolio effect helps account for the relationship between trends in high-powered money growth and real activity over 1934-39. In the fourth section, I lay out what my argument implies for relationships between yields and factors causing short-lived fluctuations in high-powered money. In the fifth section I test those implications. Finally I examine weekly series on Treasury debt issuance to distinguish portfolio effects of high-powered money shocks from changes in bond supply.

1. Current literature on portfolio effects of QE

A QE operation affects bond yields through signalling if it changes beliefs about the time when overnight rates will rise above their current low level and/or the path rates will follow thereafter. QE has portfolio-balance effects if it changes spreads between expected future overnight rates and bond yields.

Some economists (e.g. Krishnamurthy and Vissing-Jorgensen, 2011) argue portfolio effects are largely confined to a "local supply" channel (also called the "scarcity" or "market-segmentation" channel) through which changes in the supply of a bond to the public affect yields only for bonds with similar duration and default (or prepayment) risks. Others (e.g. Gagnon et. al. 2011) argue for a "duration channel" of portfolio effects. Through this channel, QE purchases of any long-term bonds could lower yields on all bonds by reducing term premiums investors require to bear duration risk. Gagnon et. al. (2011:5,7) describe the duration channel this way: QE bond purchases "reduce the supply to the private sector of assets with long

5 duration..and increase the supply of assets (bank reserves) with zero duration...With less duration risk to hold in the aggregate, the market should require a lower premium to hold that risk.” Generally, if short- term bill rates are currently zero bills are zero-duration assets equivalent to high-powered money; it should make no difference whether the central bank pays for bonds with newly created money, or by selling bills from its portfolio, effectively exchanging bonds for bills (Hamilton and Wu, 2012; Woodford, 2013).

Most arguments for portfolio effects of QE rely on the "preferred habitat" theory of Modiglini and

Sutch (1966). Much current literature (including Krishnamurthy and Vissing-Jorgensen, 2011; Gagnon et. al. 2011; D'Amico et. al. 2012; Hamilton and Wu 2012) refers specifically to a model with preferred-habitat investors developed by Vayanos and Vila (2009) to analyze financial markets with positive overnight rates.

In this model there are two types of asset: liquid zero-coupon bonds paying off at various maturities; and very short-term loans, corresponding to overnight loans, available in any quantity at an exogenously determined interest rate. The short-term rate is somewhat unpredictable so bond prices are subject to duration risk. There are two types of investor: preferred-habitat investors, each of whom demands bonds at just one maturity; and "arbitrageurs" who may hold assets in positive or negative quantities at any maturity

(bonds can be shorted). An arbitrageur is risk-averse with "mean-variance" preferences (as in Sharpe's

[1964] Capital Asset Pricing Model). The model displays a local-supply channel in that changes in bond supply at a specific maturity may have a particular effect on yields at that maturity. The model displays a duration channel in that the overall level of term premiums increases with the average duration of bonds in arbitrageurs’ portfolio. Thus a QE operation that reduces the relative supply of long-term bonds lowers term premiums, hence yields at all maturities. Along these lines, D'Amico et. al. (2012:425-26) argue that

QE works by affecting "the relative supplies of outstanding stocks of debt" hence "the average duration risk faced by investors." According to Joyce et. al. (2012:F279) "The idea here is that a central bank's purchase of long-duration assets, like medium-to-long term government bonds, will reduce the average duration of the stock of bonds held by the private sector and this may cause a fall in the premium required to hold duration risk." I refer to this type of portfolio effect as the "average-duration channel."

6 2. Portfolio effects of high-powered money: the "zero-duration channel"

In the Vayanos-Vila model there is no asset corresponding to high-powered money paying no interest. For a mean-variance investor, such an asset would be dominated by safe overnight loans. Demand for zero-interest high-powered money must be derived in other ways such as a special role of central-bank reserve balances in clearing interbank payments. In models of that role (e.g. Poole 1968, Ennis and Keister

2008), if the central bank pays no interest on reserve balances, a sufficiently large increase in high-powered money drives the return to market overnight lending down to zero, as it increases the quantity of "free reserves" (reserve balances less required minimum balances and any borrowing from the central bank). At that point, demand for high-powered money to clear payments and satisfy regulatory reserve requirements is saturated. But now money paying zero interest is not necessarily a dominated asset for a mean-variance investor, because any asset with a positive expected return is subject to duration risk, at least. A mean- variance investor could have a demand for zero-interest money, negatively related to the expected return to holding bonds (as in Tobin, 1958). To make such investors willing to hold a larger supply of outside money, the expected return to holding bonds must fall. Thus, I argue, when there is no return to safe overnight lending there may be an additional type of portfolio effect - another side of the duration channel - which I refer to as the "zero-duration channel." Through it, an increase in the supply of outside money can reduce term premiums. This is true whether or not the increase in the outside is accompanied by a change in the average duration of bond supply.

Consider a model in which, following Vayanos and Vila (2009), there are zero-coupon bonds and riskless short-term loans; preferred-habitat investors' demand for each asset depends only on exogenous factors and that asset's own interest rate or yield; and there are common beliefs about future short-term rates and bond payoffs. Following Hamilton and Wu (2012), allow for default risk on some bonds and set the model in discrete time. An arbitrageur, indexed by j, maximizes:

a (1) E∆W −j Va r( ∆ W ) tjt,1+2 jt ,1 +

7 where Wj is the arbitrageur's wealth. Following Greenwood and Vayanos (2014), allow for a negative relationship between wealth and the risk-aversion parameter aj. To simplify notation let a j= a/ W j . (I do

not rely on this specific assumption for conclusions.) A period is a day. it is the riskless overnight rate on a

daily basis . Given (1) and common beliefs, all arbitrageurs hold the same portfolio of risky bonds. rt is the

spread between it and the expected return to holding this portfolio overnight ( t to t+1). rkt is the spread

between it and the expected overnight return to holding a particular bond k. Generally:

σ k, t + 1 (2) rkt=β kt r twhe re β kt ≈ 2 σ t+1

2 σ t+1 is the perceived variance of the log of tomorrow’s value of the portfolio. σ k, t + 1 is the perceived

covariance of the log portfolio value with the log value of bond k. ( βkt is, exactly, the covariance of the realized overnight return to holding bond k with the portfolio return. The approximation holds for realistically small values of i and r.)

Unlike Vayanos and Vila, set it equal to zero: there is no return to riskless overnight lending.

Arbitrageurs expect i will remain zero for at least h days. Again to simplify notation, expectations for further days are uncomplicated: i will either remain zero for an additional day or rise above zero and remain positive within the duration of existing bonds. The probability i will remain zero on a given day beyond h is fixed, denoted π ; (1−π ) is the probability i will rise above zero on a given day beyond h assuming it has remained zero so far. Also, whatever day i happens to rise above zero, it will follow the same path subsequently. Let x denote the value for a variable x in a period before overnight rates rise

above zero, xˆυ denote the value of x in the υ th period thereafter. Then the yield to maturity on bond k is:

1dk  1  h dk − h dk − h +1 − τ  (3) i ≈Eir + = Erβ +[ πβτ r +−(1 ππ )() τ −1 irˆ + β ˆ ˆ ] kt t∑() kt+τ  t  ∑∑() kt+τ() k t + h + τ ∑ k υ  dk /ζτ =0  d k / ζ τ=0 τ = 1 υ =0 

D where dk is the bond's duration in days and ζ is the number of market days in a year. Z j is an arbitrageur’s total demand for “zero-duration assets,” that is assets that pay zero nominal return but are free 8 from default risk. These assets include zero-interest money and safe bonds paying off within h days. From

(1),

  D 1 rt 2 (4) Zjt=1−2  W jt where η t ≈ σ t +1 ηta(1+ r t ) 

(ηt is, exactly, the variance of tomorrow's portfolio value divided by the square of its expected value.) Let

M S denote the supply of outside money, zero-interest money that is a net asset for the private sector. This is the quantity of high-powered money in the economy less high-powered money created by borrowing from the central bank ("discount" borrowing) (Tobin 1965:468, 1969: 27). BS is the outside supply of safe bonds paying off within h days. Thus M S+ B S is the outside supply of zero-duration assets. Equating net demand for zero-duration assets to their outside supply determines today’s expected return to holding the bond portfolio:

S S ph  rt M ttt+ B − Z (5) 2 =aηt 1 −  (1+ rt )  W t  where W is total weath of arbitrageurs, Z ph is preferred-habitat demand for zero-duration assets.

The signalling, local-supply and average-duration channels are all present here. Signalling: (3) implies a bond's yield falls if news increases the number of days i is certain to remain zero ( h), or increases the probability i remains zero beyond h days (π ), or lowers the level of overnight rates expected to prevail after overnight rates rise above zero (iˆ ). Local supply: given the nature of preferred-habitat investors’ asset demand, a decrease in outside supply of a particular bond k reduces the weight of that bond in arbitrageurs'

portfolio, which can reduce the covariance of that bond's value with the portfolio value σ k, t + 1 ; in (2) that

reduces βk and rk , which reduces the bond's yield in (3). Average duration: because news about future overnight rates can have a bigger effect on prices of longer-duration bonds, a decrease in the portfolio's

2 average duration can decrease σ t+1 (the degree of uncertainty about tomorrow's value of the bond portfolio); in (5) that reduces r , which reduces all bonds' yields in (3).

9 S S But the zero-duration channel is also present. From (5), ∂∂rt/(M t + B t )0 < : growth in the outside supply of zero-duration assets lowers the expected return that arbitrageurs require to hold the bond portfolio overnight. Under my specific assumption about the relation between wealth and the risk-aversion parameter, what matters is the fraction of arbitrageurs' wealth they must hold in zero-duration assets. More

generally, rt falls if growth in zero-duration asset supply is rapid enough relative to growth in wealth.

Thus, a policy that boosts high-powered money growth tends to reduce yields. The effect on yields, in (3),

is stronger if high-powered money growth lowers not only rt but also expected future r (the future return to holding bonds while i remains zero): such expectations will be verified if high-powered money growth is persistent. Absent such a change in expectations, the effect of money growth on the yield of bond k is:

β (6) i/ mkt M / W a ∂kt ∂ t ≈ ()t η t dk / ζ where m is log high-powered money. The effect increases with the bond’s risk premium βk because a change in the expected overnight return to the bond portfolio has a larger effect on the return to a relatively

risky bond. The effect decreases with the bond’s duration dk , because a change in the current overnight return to holding a bond has a smaller effect on the yield to maturity of a longer-lived bond.

3. Monetary policy and interest-rate trends over the 1930s

I review 1930s history to establish the following points. There was no return to safe overnight lending over most of 1934-39. Meanwhile American monetary authorities followed policies that should have affected term premiums and bond yields through portfolio-effect mechanisms, if those mechanisms were present. Over 1934-36 and 1938-39, authorities allowed unsterilized foreign gold inflows to greatly increase high-powered money and zero-duration assets relative to wealth. After July 1936, authorities frequently tried to stabilize short-term fluctuations in Treasury yields with open-market operations.

3.1 The return to overnight lending falls to zero

In the late 1920s the U.S. had active markets in overnight lending. There were fed funds loans, "call money" loans collateralized by stocks and bonds, and interest-paying demand deposits in money-center 10 banks (Haney, Logan and Gavens 1932; Turner 1931). Call money rates usually exceeded fed funds rates, perhaps because lenders' transactions costs were higher for call money. 1 Between the October 1929 stock market crash and the trough of the Great Depression in March 1933, fed funds and call money rates were sometimes extremely low.2 In June 1933 interest rates on demand deposits in money-center banks, which

were already less than a quarter percent, were set at zero by regulatory fiat. 3

Around mid-February 1934 quoted rates for fed funds and call money settled on floors apparently determined by lenders' transactions costs. Demand for loans at these rates was only sporadic. 4 Thus, the return to safe overnight lending was effectively zero. It remained zero for the next two years: every day the quoted fed funds rate was exactly 1/8 percent with negligible borrowing. The call money rate of the New

York Stock Exchange, controlled by a cartel of New York banks, was usually pegged at exactly one percent

(Federal Reserve Board of Governors 1943: Table 121; Bradford, 1941: 445). For a few months in 1935 the banks cut the call money rate to a quarter or half percent, but they found that did not cover costs of

handling the loans and "did not call forth any significant demand for funds" (Turner, 1938:89). 5 Meanwhile

Treasury bill rates were extremely low. Figure 1 plots weekly average dealers’ bid rates for bills maturing

in three months, from February 1934 through 1939, and the auction rate for new nine-month bills, which

1 A call money lender had to manage the collateral which took time and effort (Griffiss 1923: 7, 16, 49). fed funds loans were usually uncollateralized (Turner 1931: 97). Commissions charged by loan middlemen may have been higher for call money (New York banks charged 1/2% to arrange call money loans for others [Turner 1931: 68]; the bid-ask spread charged by fed funds loan brokers was 1/4% to 3/8% [Beckhart and Smith, 1932: 42]). 2Immediately after the crash overnight rates fell as the Federal Reserve purchased Treasuries in open-market operations, cut discount rates and stopped sterilizing foreign gold inflows (Turner 1931:47). Rates rose in late 1931 when Britain went off gold and the Federal Reserve tightened to prevent gold outflows from the U.S. Rates fell again in 1932 during Federal Reserve open- market operations that boosted reserve supply (described by Hsieh and Romer 2006, Bordo and Sinha 2016): from April 1932 to October 1932 the New York stock exchange call money rate fell from 2 1/2% to 1% (Federal Reserve Board of Governors 1943: Table 121); fed funds rates fell from about 2% to 1/8% (Willis, 1957, Chart I). 3 New York Times , " Cuts Rates on Deposits, June 1, 1933; "Money Rates Harden as Funds Flow Out," July 8, 1933; "Interest Stopped on Demand Funds," June 17, 1933; ""Interest Ban Cuts Demand Deposits," June 23, 1933. 4In early February fed funds loans were "in good demand" at rates up to one percent ( Journal, "Money Rates," February 8 1934, February 9, 1934). In mid-February there was "small demand" at a rate of 1/8 ( Wall Street Journal , "Money Rates," February 16). On February 23 there were " [loans] offered by all banks, with no takers, at 1/8%" ("Money Rates," February 24, 1934). On borrowing volume see also Westerfield (1938:621), Willis (1957:11) . 5 New York Times "Pegged Call-Money Rate of 1% is Ended," April 14, 1935; "Money Rates Stay 5th Month at Lows," October 1, 1935; ""Call Money Rates Raised by Banks," October 30, 1935, "Changes in Money Rates," October 31st 1935; "Banks Here Double Call-Money Rate," January 23, 1936; Wall Street Journal October 30, 1935, "Money Rates Firm as Bankers Seek 'Living Wage'", May 9 1936, "Call and Time Money Rate Advance of 1/4% Expected Next Week." 11 were issued weekly from 1935 through 1937. The three-month rate fell below 20 basis points in March, presumably indicating the onset of near-certainty that the return to overnight lending would remain zero within a three-month horizon, at least.

Figure 1 also plots series for bond yields I rely on below. "BAA" is an average of yields on private- sector BAA-rated bonds, which several studies of the 1930s (e.g. Temin 1976, Gordon and Veitch 1986) have used to indicate business borrowing costs. “Treasury 3-5 year” is the average for medium-term

Treasury notes due in 3-5 years. “Treasuries, Long-term bonds” is the average for long-term Treasuries

(due or callable after 12 years). These three series are weekly (ending Saturday) averages of daily market- close yields. “Treasuries, Ten year” is a monthly (last market day of month) constant-maturity estimate constructed by Cecchetti (1988) to account for a peculiarity of Treasury bonds and notes in this era. In the

1930s most Treasury bond and note issues were sold at fixed prices equal to par value. To ensure an issue would sell, Treasury officials set coupon rates a bit above current market yields. To simplify refunding the

Treasury allowed holders of redeemed notes and bonds to take payment in cash or exchange them for newly issued debt. Because Treasury issues were a good deal they were usually oversubscribed and rationed, so the "exchange privilege" tied to old notes and bonds had an option value which lowered their yields. Cecchetti strips out an estimate of this option value, leaving yields comparable with later eras'. The spike in yields in August 1939 was due to the advent of the Second World War (British-Polish treaty in

August, German invasion of Poland in September).

3.2 Treasury and Federal Reserve operations to stabilize Treasury yields

At times both the Federal Reserve and the Treasury attempted to influence Treasury yields by buying or selling Treasury debt in the open market. These operations were like QE in that they were meant to influence yields by affecting the supply of debt to the public, not by affecting current or expected future overnight rates. But they were also unlike post-2008 QE in important ways. Most of the operations were deliberately kept secret. They were aimed at Treasury yields specifically, not corporate yields. They were not intended to influence long-term trends in yields but merely to keep markets “orderly” - to slow down

12 sudden increases or decreases in yields. 6 Treasury officials especially wanted to keep yields from rising just before issues of new notes and bonds. Recall notes and bonds were usually issued at fixed prices with preset coupon rates. If market yields rose just before an issue investors might not buy it. 7 But policymakers

also wanted to stop sudden decreases in yields. It is not clear why, though some policymakers seem to have believed that a sudden drop in yields was likely to be followed by a sudden rise that would hinder sales of a new issue. 8 But many yield-stabilizing operations took place when no debt issue was imminent. There were operations to stabilize Treasury bill rates, even though bills were always auctioned. 9

Federal Reserve Bank of New York staff handled operations on behalf of the Treasury as well as the

Federal Reserve system. To slow down a decrease in yields New York Federal Reserve staff sold

Treasuries through private dealers. To hinder yield increases they placed bids through dealers at prices a bit

below recent market prices so that a downtick in price triggered purchases. If they found they were doing

most of the buying in the market, they would lower their bids. The Treasury’s yield-stabilizing operations

6Information about these operations is in FOMC minutes (online at Federal Reserve Bank of St. Louis’ FRASER), papers of Treasury Secretary Henry Morgenthau (the “Morgenthau Diaries” henceforth “MD”, online at Roosevelt Presidential Library’s website) and papers of Federal Reserve Bank of New York head George Harrison (Harrison papers or “HP,” in Columbia University Rare Books library). Operational procedures are detailed in MD Vol 60:125; FOMC Minutes April 5, 1937, September 18, 1939. To hide an operation, bids and asks were placed through private dealers. Sometimes Federal Reserve staff simply gave a dealer a put for all the bonds he could get his hands on (MD Volume 60: 56 [March 1937]). In October 1937 Federal Reserve staff arranged delivery dates of purchased bonds to coincide with run-offs of Federal Reserve Treasury bill holdings so that the operation could not be inferred from the Federal Reserve's weekly statement (HP Box 24 Binder 52, Volume VII, Memorandum Meeting of Board of Directors October 21, 1937). When Federal Reserve staff wanted to keep an operation secret, they could. Operations deliberately made public were reported in the Wall Street Journal , the New York Herald Tribune and the New York Times (e.g. New York Times April 2, 1937, "Morgenthau sees 'Orderly Market'). Operations meant to be secret are not mentioned in the papers (e.g. Treasury and Federal Reserve operations of July 1936). In March 1937, the Treasury began to buy bonds on March 9th (FOMC Minutes, March 15, 1937: 5). First rumors of the operation appeared several days later (New York Times , "Treasury's Bonds Regain Some Poise," March 14, 1937). Reports became definite only on March 17 ( New York Times , "Slump is Renewed in Federal Bonds," March 17, 1937). 7 In this respect 1930s yield-stabilizing operations resembled the Federal Reserve's "even-keel" operations of the 1950s-60s, which were meant to assist Treasury debt issuance by holding yields steady. But even-keel operations were not meant to affect yields through bond supply, at least in view of Federal Reserve staff and policymakers at the time. They were instead meant to avoid disturbances to short-term money market interest rates by holding discount rates unchanged and making suitable adjustments to free reserve supply (Humpage and Mukherjee 2013: 6; Meltzer 2009: 85, 122, 157, 176, 187). In the 1930s Federal Reserve policymakers believed that changes in free reserve supply and discount rates had lost their effectiveness to influence short-term interest rates. 8 Treasury Secretary Morgenthau said "everything that goes up has to come down” (MD Volume 169:126); see also FOMC Minutes (April 22, 1938:5, April 29, 1938:5-6). Both the Treasury and Federal Reserve tried to prevent sudden decreases in yields (e.g. FOMC Minutes March 20, 1939). 9 Operations to control bill rates took place in April 1937 (FOMC April 6, 1937; April 19, 1937, April 24, 1937), and perhaps at other times. 13 were technically on behalf of Treasury-controlled agencies (the Exchange Stabilization Fund, postal savings system and "trust accounts" such as civil service retirement). In these operations high-powered money was exchanged for Treasury debt as I explain below. Federal Reserve policymakers did not want their operations to affect reserve supply (though they did not counteract reserve-supply effects of Treasury operations). Thus, in operations on behalf of the Federal Reserve, staff matched purchases at one maturity to decreases in holdings of other maturities.10 Effectively they exchanged bonds and notes for shorter-term

Treasury debt, usually bills.

Unfortunately, no data indicate the timing and magnitude of the Treasury's operations, or the

Federal Reserve’s operations prior to January 1937.11 But archival evidence shows yield-stabilizing

operations generally were rare before mid-July 1936. From 1934 through early July 1936 there was one

joint operation by the Federal Reserve and Treasury, no operation by the Federal Reserve alone, and just a

few (perhaps just two) operations by the Treasury alone.12 Starting in mid-July 1936 yield-stabilizing operations became common. At that time the Federal Reserve announced an upcoming hike in reserve requirements. Treasury officials feared the announcement would lower bond prices. They pressed the

Federal Reserve to join them in buying bonds.13 More joint Federal Reserve-Treasury operations took place

in September, October and December 1936, and January, February and March 1937, while FOMC

10 Exceptions occurred in April 1937 and November 1937 (FOMC Minutes, April 19, 1937; November 30, 1937). 11 No data distinguish between Yield-stabilizing operations and ordinary investment transactions for Treasury-controlled agencies, which were frequent. Treasury officials believed there was no way to infer their yield-stabilizing operations from published data (MD Vol. 169:133-134). FOMC minutes, MD and HP mention many operations without giving exact dates or dollar values. 12 The two Treasury operations were in August 1934 (MD Volume 2:13) and August 1935 (HP box 28, Binder 61, "Office Correspondence Harrison to Burgess,” August 27 1935). The joint operation was in March 1935 (FOMC Minutes March 13, 1935:2). Harrison, whose staff carried out all yield-stabilizing operations, told the New York Federal Reserve's Board of Directors this was the "the first time he knew of" a Federal Reserve operation (HP, Box 23, Binder 51, "Meeting of Executive Session of Board of Directors, March 21, 1935, p. 21). There is no mention of any Federal Reserve operation between March 1935 and July 1936, and two specific statements that no Federal Reserve operations were taking place in that span (FOMC Minutes, May 28, 1935; HP box 28, Binder 61, "Office Correspondence, Harrison to Burgess, August 27 1935). To each FOMC meeting prior to March 1936 HP presented a “Report of Open Market Operations.” Harrison described the March 1935 operation in the "Report" for the April 1935 FOMC meeting (as one “to assist in maintaining an orderly market”). No other “Report” mentions a Yield-stabilizing operation. 13 MD Vol. 28:98-99, July 15, 1936; HP Box 22, Binder 47, Conversation with Eccles and Morgenthau, July 15, 1936. 14 members debated the desirability of the actions. 14 In April 1937 the Federal Reserve began "two-way control," selling bonds to prevent yields from falling. 15 In May 1937 Federal Reserve yield-stabilizing operations became routine: FOMC members voted unanimously to give the executive committee broad authority to engage in open-market operations "to prevent a disorderly rise as well as a disorderly fall" in prices (FOMC Minutes, May 5, 1937: 8). From then through 1939 the authority was renewed without debate within the FOMC. The Treasury continued its own independent operations.16

If portfolio-effect mechanisms were present, yield-stabilizing operations should have worked, because they affected the supply of bonds and zero-duration assets (high-powered money and/or bills). But it is hard to observe their effects on yields because they were undertaken in response to changes in yields.

This is shown by data that indicate Federal Reserve Yield-stabilizing operations, available beginning with

January 1937. These are weekly series for total face value of Treasury debt held by the Federal Reserve in two broad maturity classes: maturing within five years, maturing after five years. Most (not all) Federal

Reserve operations to stabilize yields should appear as changes in the longer-maturity class.17 Table 1, columns (1)-(2) show results of regressing weekly changes in this class on current changes in the weekly

Treasury yield series plotted in Figure 1. (Changes in the two yield series are too strongly correlated

[coefficient 0.73] to put them together on the right-hand side.) The positive coefficients, significantly different from zero at a five percent level or better, are what one should observe if the Federal Reserve

14 MD Vol. 32:227 (September 15, 1936), Vol. 41:260 (October 27,1936); Blum 1959:358; HP Box 23, Binder 51, Memorandum on Meeting of Board of Directors December 10 1936; HP Box 22, Binder 47, note on telephone call to Eccles January 13, 1937, Box 22, Binder 47, note from W. Randolph Burgess to Harrison February 2, 1937, FOMC Minutes March 15, 1937; MD Vol. 59:9-11, 96-102, 313-14, 317-18, 321-57, Vol. 60:60-79, 80-87. 15 FOMC April 3, 1937, April 4 1937; HP Box 22 Binder 47, Memo from J.H. Williams to Harrison April 14, 1937; HP Box 22, Binder 47, Confidential Files April 10, 1937; HP Box 22 Binder 47, notes on telephone conversation with Mortgenthau April 15, 1937; FOMC Minutes April 19, 1937;HP Box 22, Binder 47, Confidential Files April 22, 1937. 16 Routine renewal of authority for Federal Reserve operations: FOMC minutes June 9, 1937, July 6, 1937, September 4, 1937, November 9, 1937, December 1, 1937, April 29, 1938, September 21, 1938, December 30, 1938, March 7, 1939. Operations: FOMC July 6, 1937; August 18, 1937; HP Box 24 Binder 52, Minutes of Meeting of Board of Directors October 21, 1937; HP Box 22, Binder 47, Confidential Files November 8 and 16, 1937; December 10, 1937; FOMC February 28, 1938;March 1, 1938; April 21, 1938; MD Vol 120: 261-264; FOMC September 15, 1938; April 25, 1938. Independent Treasury operations mentioned in FOMC Minutes March 13, 1939. 17 Some Federal Reserve operations mentioned in sources do not appear as changes in either maturity class (e.g. FOMC April 19, 1937; FOMC July 6, 1937; August 18, 1937; HP October 21, 1937; FOMC September 15, 1938; MD June 6, 1939), presumably because purchases were matched to sales within the same class. 15 indeed tended to buy (sell) long-term bonds in response to an increase (decrease) in yields.

3.3 Gold inflows, trend growth in high-powered money and economic recovery

In the late 1920s the U.S. was part of the international gold standard system. Each country's monetary authority, usually a central bank, freely exchanged currency and central bank deposits for gold at a fixed price, effectively fixing foreign exchange rates. The U.S. usually ran a balance of payments surplus which meant the Federal Reserve was purchasing gold from abroad. That tended to add to the high- powered money supply, but the Federal Reserve usually sterilized the effect by simultaneously selling bonds in open-market operations (Friedman and Schwartz 1963: 279-287; Meltzer 2003: 166-67).

In March 1933 the incoming Roosevelt administration suspended exchange of gold for dollars. In

January 1934 American monetary authorities resumed gold payment at a new, higher dollar price per ounce which devalued the dollar against currencies of countries which continued to fix or at least tightly manage their gold values.18 From 1934 through 1939 the U.S. usually ran a balance of payments surplus due mainly

(as a matter of accounting) to an inflow of international investment. Foreign demand for American dollar financial assets was conspicuously strong while American purchases and holdings of foreign-currency financial assets were minimal. The extreme imbalance of international investment has been attributed to the rise of Hitler and other European political events that threatened returns on European investments and/or gold values of European currencies.19

Under new arrangements, the Treasury, not the Federal Reserve system, exchanged dollars for gold

(buying gold from anyone, selling only to foreign monetary authorities). 20 But just as before gold purchases

18 France, Italy and the Netherlands maintained gold convertibility until September 1938. Britain held pound/gold exchange rates within tight bands: from April 1934 through August 1936 the monthly average dollar-pound rate did not vary more than four percent up or down. Most British dominions pegged to the pound. 19 Bloomfield (1950:6,11,23-27,109,114), Friedman and Schwartz (1963:474, 545), Romer (1992:773). Goldsmith (1958: 137, 145) estimated that the total value of foreign assets held in the U.S. in 1933 and 1939, including foreign assets denominated in dollars, was less than one percent of assets held by American financial intermediaries (including banks). European financial intermediaries held much larger shares of their assets in American dollar assets (e.g. Wilkins 2004: 367). 20 Meltzer (2003:458). The Treasury bought foreign gold directly and through a Treasury-controlled agency, the Exchange Stabilization Fund (ESF). The ESF sometimes bought foreign financial assets but it always converted them into gold quickly (Bloomfield, 1944: 70-71). The Treasury also purchased silver and gold from domestic producers, which contributed to 16 tended to add to high-powered money. Until December 1936 neither the Treasury nor the Federal Reserve sterilized gold purchases. The Federal Reserve took care that its own operations neither added to nor subtracted from high-powered money. Thus, high-powered money grew rapidly.

Figure 2 shows this rapid growth. The figure plots the monthly-frequency estimate of high-powered money from Friedman and Schwartz (1963) less discount borrowing from the Federal Reserve. In standard terminology this quantity is “nonborrowed base” (equal to nonborrowed reserves plus circulating currency).

From February 1934 through 1939 there was very little discount borrowing so nonborrowed base was not substantially different from high-powered money - the two quantities would be indistinguishable on this figure. 21 Nonborrowed base is the portion of high-powered money that is a net asset for the private sector

(Carlson 1981: 4). Thus, it corresponds to outside money M S in the model above. It is also the quantity of high-powered money directly created by central bank open-market operations and international transactions.

Growth in nonborrowed base greatly exceed growth in required reserves and the public's demand for currency, so excess reserves (nonborrowed reserves less required reserves) also grew. Federal Reserve policymakers came to fear that this excess reserve growth must eventually create uncontrollable inflation, as it was becoming too large to drain with sales of Federal Reserve assets. In July 1936 the Federal Reserve

Board announced a coming hike in reserve requirements to reduce excess reserves. The announcement and effective dates are marked on Figures 1 and 2 as events 1 and 1a. In January 1937 the Federal Reserve announced more hikes in reserve requirements (event 3) to become effective the following March (3a) and

May (3b). Federal Reserve officials believed, and told the public, that these hikes would have no immediate effect on financial conditions (Goldenweiser 1951: 175-82; Federal Reserve Board Press Release July 14,

1936). Indeed, neither announcement of reserve-requirement hikes was associated with an unusual increase

nonborrowed base as did foreign specie purchases. But domestic specie purchases were much smaller and steadier than foreign purchases (Federal Reserve Board 1943: Table 156; Friedman and Schwartz 1963:488). 21 The absence of discount borrowing over 1934-39 is consistent with standard models of reserve demand and overnight-rate determination following Poole (1968), as noted by Hanes (2006). 17 in yields. 22

Meanwhile Treasury policymakers had come to fear foreign gold inflows might reverse and create uncontrollable deflation (Blum, 1959: 359-60). To head this off, they developed a method by which the

Treasury could sterilize gold inflows and build up a gold reserve to cover a future balance-of-payments deficit. The scheme was simultaneously announced and put into effect in the third week of December 1936

(event 2). Trend growth in noborrowed base ceased. Like the reserve-requirement hikes, the sterilization scheme was not intended to have an immediate effect on financial conditions (Blum, 1959:367). But almost immediately, yields on relatively short-term Treasury debt began to rise (Blum 1959: 367; Morgenthau

Diaries, Vol. 55: 115-117). This is indicated in Figure 1 by upturns in rates on nine-month bills and 3-5 year notes.

In March 1937 a previously-announced reserve-requirement hike became effective (event 3a), greatly reducing excess reserves. At ths time the return to overnight lending appeats to have risen above zero: demand for fed funds loans revived; quoted fed funds rates rose above the floor where they had been since February 1934. At the same time the three-month Treasury bill rate rose above 20 basis points for the first time in three years; long-term Treasury yields began to rise (Figure 1). Over the next few months fed funds rates were sometimes as high as half a percent. 23

In June 1937 came the downturn of the 1937-38 recession (NBER cyclical peak in May, marked "P" in Figures 1 and 2). Policymakers perceived that recovery had stalled (FOMC minutes, June 9 1937). They began to reverse course. In September 1937 it was announced (event 4) that the Treasury would release some sterilized gold into the money supply (Blum 1959:377-79). Money growth resumed (Figure 2). In late

22 Using a standard event-study method (following Swanson, 2011) I checked whether either announcement date (July 15, 1936, January 31, 1937) saw a statistically significant change in weekly yield series, using standard deviations of weekly changes in the series over a span of time prior to the announcements (March 1934- July 1, 1936). Neither was associated with a significant change in any yield series, even at the ten percent level. 23 Wall Street Journal March 12, 1937, "Higher Rate on Federal Funds," March 16, "Commercial Borrowing Gains Expand, Goes to 1/4%." New York Times March 12, 1937, "Topics in Wall Street." Wall Street Journal May 3, 1937, "Federal Funds Rise to 1/2%," August 6 1937 "Member Banks' Excess Reserve Touch New Low." New York Times May 4, 1937, "Topics in Wall Street" September 1, 1937, "Reserve-rate cut explained by Bank." Federal Reserve Bank of New York Monthly Review September 1, 1937:65-66. Federal Reserve Bulletin September 1937:820. 18 September fed funds rates fell to a floor again, where they remained through 1939. 24 In October 1937 three- month bill rates fell back below 20 basis points. In February 1938 (event 5) the Treasury announced further limits on sterilization (Johnson 1939:140). In April 1938 (event 6) Roosevelt and the Treasury announced that reserve requirements would be reduced immediately, all sterilized gold would be released into the money supply, and no future gold inflows would be sterilized (Blum, 1959: 425; New York Times , April 19,

1938, “Treasury Gives Up Sterilization Policy”). The April 1938 policy announcement was associated with an unusual drop in bond yields. 25 In July 1938 employment and output began to grow again (NBER trough in June, marked "T" in Figures 1 and 2).

There is an obvious correlation between trend high-powered money growth and recovery of real activity over 1934-39. As high-powered money grew over 1934-36 and 1938-39, so did employment and output. (Real GDP grew at an average annual rate of 10.9 percent 1933-1936, 8.1 percent 1938-1939

[Carter et. al. 2006: Ca6]). The downturn of the 1937 recession came about five months after money growth ceased in December 1936 (annual real GDP grew just 0.9 percent from 1936 to 1938).

Several economists have argued that high-powered money growth, or policies resulting in high- powered money growth, boosted recovery over 1934-36 and 1938-39, while cessation of growth at the end of 1936 was partly responsible for the 1937 recession. They do not argue that money growth affected nominal interest rates, because short-term rates appear to have been at or near a lower bound. Instead, like

Romer (1992), they hypothesize that high-powered money growth was associated with increases in expected future inflation which lowered long-term real interest rates.26 Bernanke, Reinhart and Sack

(2004:18-19) describe post-1933 high-powered money growth as "a successful application of quantitative easing" that boosted expected inflation by signalling a change in the monetary regime (see also Svensson

24 Federal Reserve Bank of New York Monthly Review November 1 1937: 32. From late September 1937 through 1939, every newspaper report on fed funds lending quoted the going rate as 1/4 percent. I have found no explanation of why this floor was higher than the earlier floor of 1/8. 25 The April announcement was coincident with decreases in weekly yield series that were statistically significant at the five percent level, on the test described in footnote 20. 19 2004: 90, Krugman 1998: 161).27 Eggertsson and Pugsley (2006) and Eggertsson (2008) show how 1930s

monetary policies could affect real activity through expected inflation in a standard New Keynesian model.

3.4 Trend money growth, zero-duration asset supply and bond yields

Whether or not high-powered money growth boosted expected inflation, it was certainly associated with declines in long-term nominal interest rates. 28 Cecchetti's ten-year Treasury yield fell about one percent from April 1934 to December 1936, and again about one percent from September 1937 to July

1939. The path of BAA yields paralleled Treasuries', apart from a widening in the BAA-Treasury spread from summer 1937 through spring 1938 which can be explained by rising default-risk premiums in the recession. 29 Models of the post-2008 American economy suggest these changes in nominal rates were large

enough to have substantial real effects. In several models, with short-term rates stuck at a lower bound, a

26 Other economists have argued that high-powered money growth and changes in reserve requirements affected the money stock (monetary aggregates) through the usual money-multiplier mechanism (accounting for banks' demand for excess reserves), which in turn affected current real activity (e.g. Friedman and Schwartz 1963:499-545; Brunner and Meltzer 1964: 28-32). 27 In the context of QE policies in general (not the 1930s US), Bernanke, Reinhart and Sack (2004) also discuss the possibility that QE reduces term premiums. 28 Several studies have observed correlations between growth in monetary quantities and changes in interest rates over 1934-39. Basile, Landon-Lane and Rockoff (2011) observe a correlation between bond yields and growth in the M2 moneary aggregate. Irwin (2012) observes that cessation of money growth due to sterilization was coincident with increases in interest rates. Many economists have argued that generally demand for monetary aggregates or high-powered money may be related to long-term, as well as short-term interest rates (Nelson 2003). Meltzer (1998) argues that such a relationship held in the 1930s as well as other eras.Hanes (2006) shows that weekly growth in nonborrowed reserves was negatively related to weekly changes in Treasury yields over 1934-39 as a whole and within 1934-36 specifically. Hanes presents a model consistent with this finding that focuses on banks' demand for reserves, taking as given other demands for high-powered money. In the model, a bank holds reserves and bonds and can also borrow or lend overnight. A bank maximizes the expected value of profit subject to a cost of failing to meet a reserve requirement at the end of daily settlement, and a separate cost of falling short of capital. At times when the overnight rate is positive, the model is equivalent to standard reserve-demand models following Poole (1968): a bank's demand for free reserves reflects a tradeoff of the overnight rate - the opportunity cost of holding reserves - against the benefit of holding reserves to reduce the danger of an end-of-day reserve shortfall; an increase in free reserve supply lowers the overnight rate but does not substantially affect term premiums. If free reserve supply is great enough to drive the overnight rate down to zero, the opportunity cost of holding reserves becomes the expected return to holding bonds, while the benefit of holding reserves is a reduction in the probability of a capital shortfall which could otherwise result from an unpredictable decrease in the price of bonds (duration risk). An increase in reserve supply now has a substantial effect on term premiums, and hence can reduce bond yields holding fixed expected future overnight rates. The Hanes (2006) model resembles the one presented here in section 2 and Appendix 2 in that duration risk plays a key role, and in the implication that an increase in high-powered money tends to reduce term premiums. But the model presented here has several further implications, including a larger effect of high-powered money supply on yields of riskier bonds, different relationships between yields and changes in Treasury Money versus changes in Gold, and possible effects of changes in bond supply on yields. Hanes (2006) does not look for those patterns in the data. Also, Hanes (2006) does not account for the presence of yield-stabilizing operations, and does not compare 1934-36 with later periods in which yield-stabilizing operations were common. 29 The BAA default rate increased over 1936-38, as it had over 1930-1933 (Moody's Investors Service 2011:27) when the BAA- Treasury spread had widened due to default-risk premiums (Temin 1976: 103-121). 20 compression of term premiums that reduces the ten-yield nominal Treasury yield by one percent - the decline that actually occurred over 1934-36 and 1937-39 - boosts real GDP growth by more than two percent (Fuhrer and Olivei, 2011; Chung et. al. 2012; Baumeister and Benati 2013). Certainly, these declines in yields were large relative to estimated effects of post-2008 QE operations. According to the most generous estimates, the total effect of all post-2008 operations was to reduce ten-year Treasury yields by about one percent (Greenlaw et. al. 2018).

The average-duration channel cannot explain movements in long-term yields over 1934-39.

Through this channel, yields could fall as a result of a decrease in the average maturity of Treasury debt supplied to the public (Hamilton and Wu, 2012; Greenwood and Vayanos, 2014). But available data, summarized in Table 2, indicate average maturity actually increased as yields fell over 1934-36 and 1938-

39, and decreased as yields rose over 1937. Column (1) gives an estimate of average maturity of Treasury debt outstanding (issued, not yet redeemed) excluding bills which may have been zero-duration assets.

(Trends were similar including bills). Column (2) gives, for debt outstanding less Federal Reserve holdings, the percent (face value) with maturity greater than five years.30 Both (1) and (2) show upward trends in average maturity over 1934-36 and 1938-39. Column (3) shows the ratio of Treasury debt held by the public to GNP, discussed by Greenwood and Vayanos (2014) and Krishnamurthy and Vissing-Jorgensen

(2012). It was not trending down either.

The zero-duration channel can explain movements in yields over 1934-39. When there is no return to overnight lending, as from March 1934 through 1936 and from September 1937 through 1939, growth in the outside supply of zero-duration assets relative to wealth should reduce term premiums. Zero-duration assets may have included Treasury bills, at least relatively short-dated ones. 31 But the sum of nonborrowed base and Treasury bills held by the public, plotted in Figure 2, followed a path similar to nonborrwed base

30 I have found no data on maturities of Treasury notes and bonds held by Treasury-controlled agencies, but there is no reason to believe it was very different from maturities of debt outstanding. 31 Zero-duration assets would not include short-dated Treasury bonds, notes or certificates. These came with the exchange- privilege option and hence were subject to a type of duration risk: the option’s realized value depended on what long-term yields turned out to be at the time of new debt was issued - if yields rose unexpectedly the option was worth less. 21 alone. From March 1934 to December 1936 average annual growth in nonborrowed base or nonborrowed base plus bills was about 15 percent. From October 1937 to July 1939 nonborrowed base grew at an annual average rate of 17 percent; nonborrowed base plus bills at 12 percent. Wealth was growing at the same time but not nearly as fast, according to available estimates. 32 Figures 3 and 4 scatter Cecchetti’s ten-year

Treasury yield against log nonborrowed base, and the sum of nonborrowed base and Treasury bills. Both figures show a strong negative relationship, except from December 1936 through April 1938 during which fed funds and bill rates briefly rose above their floors and there were several big policy announcements.

The zero-duration channel is also consistent with the interest-rate response to cessation of nonborrowed base growth in December 1936. Recall rates on short-term Treasury debt began to rise immediately; overnight rates, three-month bill rates and very long-term yields did not rise until the following March. Cessation of nonborrowed base growth meant zero-duration asset supply was falling relative to wealth in January 1937 (wealth must have continued to grow through early 1937 - real activity and stock prices were still rising). A resulting increase in the current overnight return to holding bonds, prior to any change in expected future overnight rates, would most strongly affect yields on relatively short- duration debt.

However, the correlation between yields and trend growth in nonborrowed base or high-powered money might also be due to the signalling channel. Perceived changes in trend growth could affect expectations about the future point in time overnight rates would rise above their floor and the path they would follow subsequently.

4. What the zero-duration channel predicts for week-to-week fluctuations

To test whether the zero-duration channel of portfolio effects was indeed present in the 1930s as I

32 On the annual estimates of Ando and Brown (1963, Appendix Table I-AI, column 4), based on slim information, American household wealth grew at an annual average rate of about 5 percent from January 1934 to January 1937, about 3 percent from January 1938 to January 1940. Goldsmith's estimates, based on more information, are only for 1933 and 1939. Between those years, average annual growth rates for net worth of nonfarm households, banks, and all private financial intermediaries including banks were respectively 3.25, 1.69 and 1.43 percent. (Nonfarm households is from Goldsmith 1956, Wealth of nonfarm households, line IV, Table W-13 and W-14. Banks is Goldsmith 1958, Total Banking System Appendix Table A-1, minus net 22 hypothesize, I look for effects of shocks to zero-duration asset supply that could not affect yields through the signalling channel because they were essentially transitory, without implications for future high- powered money supply or overnight rates. Shocks of this type were created by fluctuations in weekly flows of payments to and from the Treasury’s Federal Reserve accounts. These shocks should have affected bond yields through the zero-duration channel except at times when the Federal Reserve and Treasury were engaging in yield-stabilizing operations. Thus, these shocks should be observed to affect yields from 1934 through early July 1936, not so much after that. In this section, I describe the relationship between zero- duration asset supply, Treasury payments and other factors affecting nonborrowed base. I lay out implications of the zero-duration channel for the relationship between these factors and short-term fluctuations in yields. In the following section, I will test these implications.

4.1 Factors causing short-run shocks to the supply of money and zero-duration assets

Nonborrowed base was (ignoring small factors) equal to currency left over from the pre-Federal

Reserve era plus dollars created by specie purchases and Federal Reserve open-market operations minus dollars held by U.S. or foreign government authorities, withdrawn from supply to the public. Figure 5 plots data on factors affecting nonborrowed base at the shortest available frequency, weekly (Wednesday), along with a weekly estimate of nonborrowed base scaled to match the monthly estimate in Figure 2. "Gold" is

Treasury gold holdings valued at the fixed dollar price: increases in this item indicate gold purchases adding to nonborrowed base. "Federal Reserve securities portfolio" is the par value of the Federal Reserve's

Treasury securities: increases here indicate Federal Reserve open-market operations adding to nonborrowed base. "Treasury Money" is dollars held by the Treasury and Treasury-controlled agencies in Federal

Reserve accounts and other ways: increases in this item subtracted from nonborrowed base.33 "Other" is the

worth of Federal Reserve Banks Table A-2. Financial intermediaries is “Financial Interrmediaries” Table 27 minus Federal Reserve Banks [Table A-2], Federal Land Banks [A-23] and Government lending institutions [A-24].) 33 Treasury Money also includes stocks of currency and outside circulation and "gold in the general fund.” A gold purchase added to nonborrowed base only after the Treasury created “gold certificates” (or “gold certificate credits”) “backed” by the purchased gold, "deposited" the gold certificates in a Federal Reserve account and spent the funds out of the account. Purchased gold that had not yet been deposited at the Federal Reserve appeared in Treasury accounts as "gold in the general fund" (Federal Reserve Bulletin April 1935:419-427). 23 net contribution from remaining factors. To sterilize gold purchases starting in December 1936, the

Treasury created procedures that automatically matched gold purchases to equal increases in Treasury

Money. To de-sterilize the Treasury reduced Treasury Money by spending funds out of its Federal Reserve accounts. In Figure 5 these operations appear as a buildup, plateau and drawdown in Treasury Money. 34

Across most weeks the change in nonborrowed base was practically equal to the change in Gold minus the change in Treasury Money, because Federal Reserve open-market operations had very little effect on nonborrowed base over 1934-39. Table 3, panel A shows results of regressing weekly changes in nonborrowed base on changes in Gold and Treasury Money. Pre- and post-sterilization, the R2 is 0.99.

(During sterilization the coefficients on Gold and Treasury Money are still practically one - sterilization operated by counteracting increases in Gold with equal increases in Treasury Money - but the R2 is lower because there were some Federal Reserve open-market operations in this period.) What about the total supply of outside zero-duration assets, perhaps including Treasury bills? It is hard to define exactly which maturities of bills could have been zero-duration assets at which points in time, but it is clear that changes in Gold and Treasury Money must have affected the supply of zero-duration assets much as they affected nonborrowed base alone. From Treasury records I constructed a weekly (Wednesday) estimate of bills outstanding (Appendix 1). Table 3, panel B shows results of regressing weekly growth in the sum of bills outstanding and nonborrowed base on growth of Treasury Money and Gold. Pre- and post-sterilization, estimated coefficients on Treasury Money are close to those for nonborrowed base alone in panel A and the

R2 is above 0.90.

Weekly gold purchases reflected the balance of international payments, hence international investment flows. Contemporary observers and later economics literature attributed fluctuations in the rate of gold purchase to political events that spurred or slowed capital flight from Europe (Friedman and

Schwartz 1963:509, Federal Reserve Board 1936:8-10; Federal Reserve Bulletin May 1936:311-12). To the

34 To sterilize the Treasury created a new accounting item equivalent to “gold in the general fund,” and simply refrained from creating gold certificates, letting “inactive gold” build up. To desterilize the Treasury converted inactive gold into funds in 24 degree that was true, fluctuations in Gold growth were exogenous to weekly changes in American financial conditions. But it is plausible that foreign demand for American assets, hence Gold growth, was also affected by expected returns to holding American bonds. In any case, increases in Gold were rarely reversed. Table 4 gives serial correlation coefficients for weekly growth in Gold and Treasury Money at one, two and three-week lags. For Gold the coefficients are positive. Large weekly Gold growth meant a persistent increase in nonborrwed base, with further growth likely in upcoming weeks.

Weekly changes in Treasury Money, outside the sterilization period, were mainly due to fluctuations in the flow of payments into and out of the Treasury’s Federal Reserve accounts. In the 1930s nearly all payments between the public and the Treasury or its agencies were made through these accounts, including payments associated with the Treasury’s yield-stabilizing operations. Thus, purchases (sales) of notes and bonds in the Treasury’s yield-stabilizing operations added to (subtracted from) nonborrowed base. Unlike weekly gold purchases, effects of weekly changes in Treasury Money on nonborrowed base were essentially transitory, tending to reverse in subsequent weeks. Contemporaries understood this

(Federal Reserve Bulletin July 1935: 426; Federal Reserve Board 1939:65). In Table 4, serial correlation coefficients for Treasury Money growth are all negative pre- and post- sterilization.

Prior to July 1936, while yield-stabilizing operations were rare, payments out of Treasury accounts must have been predominantly for debt coupons and principal and ordinary government expenditures; payments-in, from tax revenue and cash sales of new debt. Changes in Treasury Money were especially large in weeks around quarterly tax payments and debt issues (Burgess 1936 chapter 7, Federal Reserve

Bulletin April 1935: 201-202, Simmons 1940). Dates and amounts of debt issues were set in advance. Thus, most weekly changes in Treasury Money must have been exogenous to short-term fluctuations in financial conditions. The weekly change in Treasury Money was a negative shock to nonborrowed base and the supply of zero-duration assets. If the zero-duration channel was present, one should observe a positive relationship between weekly changes in Treasury Money and changes in bond yields.

Federal Reserve accounts and spent the funds. The buildup in Treasury Money in early 1936 was funds for the June 1936 25 This positive relationship should be weaker or absent in periods after July 1936 when yield- stabilizing operations had become common. Presumably the Treasury’s yield-stabilizing operations (recall no data indicate these) responded to current changes in Treasury yields as did the Federal Reserve’s operations (Table 1). Thus the Treasury’s yield-stabilising operations would tend to create a negative relationship between weekly changes in yields and Treasury Money: bond purchases (sales) in response to increases (decreases) in Treasury yields reduced (added to) Treasury Money. To the degree that changes in

Treasury Money were still due to exogenous factors after July 1936, any potential effect on yields could be counteracted by the Federal Reserve's yield-stabilizing operations. When the FOMC adopted a standing policy to stabilize yields, committee members specifically mentioned "preventing disorderly conditions over a quarterly income tax payment period or Treasury financing period" (FOMC Minutes, May 5, 1937:

3). Data on Federal Reserve holdings of long-term Treasuries confirm that Federal Reserve staff used yield- stabilizing operations in this way. Table 1, column (3) shows results of regressing weekly changes in

Federal Reserve long-maturity holdings on growth in Treasury Money and Gold, starting at the end of April

1938 when sterilization (which had special effects on Treasury Money) ceased. The positive coefficient on

Treasury Money, significantly different from zero at the one percent level, is consistent with a hypothesis that the Federal Reserve bought (sold) Treasury bonds when there were increases (decreases) in Treasury

Money that would otherwise have tended to increase (decrease) yields.

I return to the model above to lay out implications for relationships between weekly changes in yields, Treasury Money and Gold. I add features to the model to match 1930s conditions. To describe links between high-powered money, gold inflow and foreign demand for American assets, I add a foreign currency, foreign bonds and overnight loans that pay off in that currency, foreign arbitrageurs who maximize expression (1) in terms of that currency, and an uncertain exchange rate between that currency and dollars. The change in the American high-powered money supply from one period to the next is the change in Gold minus the change in Treasury Money. In the model, changes in Treasury Money are

veteran's bonus payment ( Federal Reserve Bulletin July 1936: 509). 26 transitory, consistent with the conditions of 1934-36/1938-39, and do not affect expectations of future returns to bondholding or overnight rates. Changes in Gold can reflect both exogenous factors such as changes in the perceived probability of foreign-currency depreciation, and an endogenous response of international asset demand to expected returns on American bonds. I allow for a relationship between changes in Gold and expected future bond returns because gold growth was associated with persistent, continuing increases in high-powered money.

Results from this extension of the model are intuitive. I summarize them here and give details in

Appendix 2. I consider potential results of regressing period-to-period changes in yields on changes in log

Treasury Money and log Gold, each scaled to the magnitude of its effect on log high-powered money. In a sample where yield-stabilizing operations are rare the estimated Treasury Money coefficient is positive.

The magnitude of the coefficient for a particular bond increases with the ratio of the bond's risk premium to

its duration ( βk / dk ) as in expression (6). In a sample where yield-stabilizing operations are common the

Treasury Money coefficient is biased toward zero. In either type of sample, the sign of the estimated Gold coefficient is ambiguous. It depends on the relative importance of exogenous factors affecting foreign demand for American assets. If such factors are relatively unimportant, the Gold coefficient can be positive: gold inflows are associated with increases in expected returns to American bonds which draw in foreign investment. On the other hand, if exogenous factors are relatively important the Gold coefficient is negative: most Gold inflows represent exogenous increases in the money supply. The absolute magnitude of the Gold coefficient can be greater than the Treasury Money coefficient because increases in Gold can be associated with downward revisions to expected future bond returns and/or future overnight rates.

5. Tests

To test these implications I regress first differences in the three weekly yield series - Treasury 3-5 year notes, Treasury long-term bonds, BAA bonds - on weekly changes in log Treasury Money ∆tres and log Gold ∆g scaled to their effects on log nonborrowed base, that is (Tres / M ) ∆ tres and (G / M ) ∆ g . I compare results from a sample running from March 1934 through early July 1936, when the return to 27 overnight lending was apparently zero and yield-stabilizing operations were rare, with results from later periods when the return to overnight lending was also apparently zero but yield-stabilizing operations were routine. I find that the 1934-36 sample gives positive, statistically significant coefficients on Treasury

Money growth, negative coefficients on Gold growth. Later samples do not. The magnitude of the Treasury

Money coefficient for a particular yield series increases with the ratio of risk premiums to duration for bonds in that series. Finally, I test the possibility that observed relationships between yields and Treasury

Money growth over 1934-36 were due to changes in bond supply correlated with changes in Treasury

Money, rather than changes in zero-duration asset supply. To do this, I use newly constructed weekly series on Treasury debt issuance.

5.1 Specific implications for the three yield series

In the model the estimated Treasury Money coefficient for a zero-coupon bond increases with the

ratio of the bond's risk premium ( βk ) to its duration( dk ), absent yield-stabilizing operations. What does this imply for the three yield series, which are average yields to maturity of coupon bonds? Assuming a coupon bond was valued as a package of implicitly priced zero-coupon bonds, its estimated Treasury

Money coefficient should increase with the ratio of the bond’s risk premium to (Macaulay's) duration (the average duration of its payments weighted by the contribution of each payment to the current price of the bond). That means the coefficient on Treasury Money should be larger for Treasury 3-5 year notes than for long-term Treasury bonds: over 1934-early July 1936 average duration of the notes was less than one-fifth of the bonds' while relative yields indicate the average risk premium on the bonds was less than twice that on the notes.35 The coefficient for BAA bonds should also be larger than the coefficient for long-term

35 Securities in the Treasury series are given by Federal Reserve Board (1943: 427-429), Federal Reserve Bulletin (May 1936:317-319; December 1938:1045-46) and Bulletin of the Treasury Department (July 1941: 21-24). U..S. Treasury Department (1934, Table 17; 1935, Table 17; 1936, Table 18) gives coupon rates, dates of maturity and first possible call for Treasury debt outstanding on June 30th of each year. For securities in each series I calculated duration to maturity and to first possible call (approximating yields by the value of the corresponding average yield series in the week of June 30th). For securities in the 3-5 year note series, which were not callable, average duration to maturity was 1.6 years (1934), 3.4 (1935), 2.7 (1936). For securities in the long-term bond series (first callable 3, 4 or 5 years before maturity depending on the bond), average duration to maturity/ first possible call was 14.5/12 (1934), 14/13(1935), 15/13(1936). Going off duration to first possible call, the average ratio of durations for bonds/notes 1934-1936 was thus at least 5.36. An upper bound on the ratio of risk premiums for 28 Treasury bonds, because the BAAs were riskier than the long-term Treasuries and somewhat shorter in duration. 36

5.2 Results for bond yields

Table 5 shows regression results. I added quadratic time-trend terms to the right-hand side to make sure apparent relationships were not due merely to correlated trends. Results were similar if I excluded those terms, or added a variable measuring the effect on nonborrowed base of Federal Reserve open-market operations (which was practically zero in most weeks.)

For panel A the sample runs from the beginning of March 1934 through the week prior to the July

1936 announcement of reserve-requirement hikes. For Treasury notes the Treasury Money coefficient is positive and significantly different from zero at the one percent level. The coefficient’s size implies a one

SD increase in log Treasury Money from one Wednesday to the next was associated with a 1.25 bp increase in weekly average Treasury note yields. For comparison, in this sample a one SD change in weekly average note yields was 6 bp. For Treasury bonds the Treasury Money coefficient is smaller in magnitude, not significantly different from zero. For BAA bonds the coefficient is larger than for Treasury bonds and significantly different from zero at two percent. Coefficients on Gold growth (G / M ) ∆ g are negative, significantly different from zero at conventional levels and larger in magnitude than coefficients on

(Tres / M ) ∆ tres . That is what the model predicts for a sample in which fluctuations in Gold inflows were due mainly to factors exogenous to American financial markets.

bonds/notes can be inferred from the ratio of average yields, which was 2.8/1.4 percent over 1934-early July 1936. This is an upper bound because a portion of the bond-note yield spread must have been due to higher expected future overnight rates over the longer maturity horizon of bonds. 36 Average BAA yields (5.6 percent 1934-36) were more than twice long-term Treasuries' (2.8 percent). The BAA risk premium was probably not as large as this ratio suggests; the yield spread must have partly reflected Treasuries' exchange privilege and tax advantages (Daglish and Moore 2013). I estimate the BAAs' average duration to maturity from information in the "Bond Section" of Moody's Investment Survey, which occasionally listed the bonds in the series, their maturity years and coupon rates (February 19 1934: 1455; April 29 1935: 1398; October 7 1935: 1236). Again I approximate yields by the value of the BAA yield series. Their maturity was 12-13 years , just bit shorter than average duration in the Treasury bond series, but there was a big difference in duration to possible call. Nearly all corporate bonds were callable within just four years of issuance issuance (Hickman, 1960: Table 114). In the 1920s-1930s most corporate bond issues were called (Hickman 1960: Table 164). For large corporate issues over 1930-39, median years from offering to maturity was 22.2; due to calls the median actual life span of an offering was only 7 years (Hickman (1953:76-77). Investors were highly conscious of the possibility corporate bonds would be called (e.g. Moody's Investment Survey "Bond Section" March 4 1935: 1448). 29 Panels B) and C) show results from samples in which the return to overnight lending was again apparently zero, but yield-stabilizing operations had become common. B) starts just after Roosevelt's April

1938 announcement that sterilization would be abandoned. This sample is entirely post-sterilization so it is most comparable to 1934-36 apart from yield-stabilizing operations. C) starts at the beginning of October

1937, excluding weeks around the April 1938 announcement (which appears to have affected bond yields).

This sample includes weeks when gold inflows were sterilized or partially sterilized. Results from both samples differ from 1934-36: coefficients on the Treasury Money variable are not significantly different from zero. Gold coefficients are positive.

5.3. Was it really bond supply?

Changes in Treasury Money were especially large around Treasury note and bond issues, as noted above, so they were correlated with changes in debt supply. Was the apparent relationship between changes in yields and Treasury Money within 1934-36 entirely due to portfolio effects of correlated changes in debt supply, rather than portfolio effects of high-powered money? A bond sale might raise yields through the average-duration channel by lengthening the average maturity of outstanding Treasury debt. A note sale did not lengthen average duration (the five-year maturity of notes was shorter than average for outstanding

Treasury debt [Table 1]), but it might raise note yields through the local-supply channel. To examine this I constructed weekly series for Treasury note and bond sales (Appendix 1). Comparing relationships between debt sales, changes in Treasury Money and yields I find evidence for a distinct effect of high-powered money, from two features of Treasury debt issuance in the 1930s.

First, payments to the Treasury for some notes and bonds did not go directly to Treasury Federal

Reserve accounts. A bank purchasing for its own holdings could pay by creating (or adding to) a demand deposit due to the Treasury. Eventually the Treasury transferred these commercial bank balances to its

Federal Reserve accounts. (It did not make payments directly out of these accounts [Simmons 1940:327].)

But in the meantime there was no effect on Treasury Money or high-powered money. If the apparent relationship between yields and growth in Treasury Money is an artifact of an underlying effect of debt

30 sales, a similar relationship should exist between yields and growth in commercial bank balances due to the

Treasury. It does not.

Second, the Treasury allowed holders of maturing debt to take payment directly in newly-issued debt - the "exchange privilege" mentioned above. Over 1934-36, most (about 60 percent) sales of Treasury notes and bonds were for "exchange," not "cash." Exchange sales affected average maturity and local supply but were relatively weakly correlated with changes in Treasury Money. I find that exchange sales appear unrelated to changes in yields. Yields are correlated with a particular type of debt sale - cash versus exchange, bonds versus notes - to the degree that type of sale is correlated with changes in Treasury

Money, not to the degree it could affect average duration or local supply.

Table 6 shows correlations between debt sale types, weekly changes in Treasury Money and changes in commercial bank balances due to the Treasury. Note debt sales are about as strongly correlated with Treasury commercial bank balances as with Treasury Money. Correlations with Treasury Money are weaker for exchange sales than cash sales, and weaker for notes than for bonds. Cash note sales are strongly correlated with commercial bank balances, indicating that banks bought a relatively large portion of notes for their own holdings. That makes sense: in the 1930s banks held a relatively large portion of the

Treasury note supply ( Treasury Bulletin April 1940:26).

Table 7, panel A) shows results of regressions like Table 5's, replacing the change in Treasury

Money with weekly debt sales. I omit results for Treasury long-term bond yields, which did not show statistically significant relationships with any type of debt sale. I exclude time trends to save degrees of freedom; adding them back made little difference to the results. Results were also similar if I replaced debt sales with the ratio of sales to outstanding Treasury debt or GNP (using the data for Table 2). For note yields, the statistical significance of the debt-sale coefficient is greater (lower p-value) as a debt sale type is more strongly correlated with Treasury Money (in Table 6). For BAA yields, no debt sale coefficient is significantly different from zero at conventional levels. Panel B) shows results of regressions with both debt sales and Treasury Money growth on the right-hand side. For note yields, Treasury Money coefficients

31 are not always significantly different from zero at conventional levels. But their statistical significance is higher (lower p-values) when the debt sale variable is weakly correlated with Treasury Money. For BAA yields the relationship is clearly with Treasury Money, not debt sales.

Finally, Table 8 shows results of regressing yields on changes in Treasury Money versus Treasury

Commercial Deposits (" ∆TresComm "). To clarify the comparison I use the change in Treasury Money rather than the scaled change in the log. Clearly yields are related to Treasury Money, not to Treasury

Commercial Deposits.

Conclusion

From March 1934 through early July 1936 week-to-week changes in yields of Treasury and corporate bonds were correlated with high-powered money shocks due to weekly fluctuations in the flow of payments to the Treasury's Federal Reserve accounts. This pattern cannot be explained by the signalling channel because these high-powered money shocks were transitory; they signalled nothing about future overnight rates. The pattern does not appear to be due to changes in bond supply associated with the high- powered money shocks. Thus, it is evidence that, when there is no return to safe overnight lending, changes in high-powered money can affect medium- and long-term yields through portfolio effects (the zero- duration channel). In periods subsequent to July 1936 the correlation between these high-powered money shocks and yields is weak or absent. This shows that yield-stabilizing operations by the Treasury and

Federal Reserve, which became common after July 1936, were effective, again through portfolio effects.

Thus I have found relatively unambiguous evidence for the existence of portfolio effects. Because increases in high-powered money supply could lower bond yields by reducing term premiums, rapid trend high- powered money growth over 1934-36 and 1938-39 reduced long-term nominal interest rates. This helps explain why trend high-powered money growth was associated with recovery in real activity over 1934-39.

Appendix 1 Newly constructed weekly data Treasury Bills Outstanding. From figures published in the Daily Statement of the Treasury.

For the last day of each month the Statement gives total par value of issued Treasury bills that were not yet

32 due, and par value value of outstanding bills that had already come due but had not been presented for payment. For each day the Statement gives revenue from sales of new Treasury bills, and expenditures for payments on Treasury bills coming due and old bills finally presented for payment, so far that month.

Because discount rates were so low, sales revenue was practically equal to the par value of newly issued bills. My figure for Treasury bills outstanding Wednesday is:

Current + past due bills outstanding on the last day of the previous month

+ revenue from bills sales so far in the current month

- expenditures on payments for currently-due and past-due bills so far in the current month.

Treasury deposits in commercial banks. From Daily Statement , Wednesday (or day preceding Wednesday holiday): "Special depositaries account of sales of Treasury bonds, Treasury notes."

Treasury Debt Sales. The Daily Statement gives cumulative sales revenues from the first day of a month to each business day. I exclude series not sold to the public (e.g. special series that went straight to Treasury- controlled agencies). Starting November 1936 the Daily Statement gives cash and exchange revenues separately; before that, only the total. To distinguish cash from exchange sales up to November 1936 I used the Treasury's Annual Reports for 1935 and 1936 which give how much of each sale was for exchange versus cash ( Annual Report for 1935: 18-18; for 1936: 12-17). The fiscal years began in July, so that leaves the weeks from April 1934 through June 1934. Within this period there were two issues, of bonds alone in

April, and of bonds and notes in June. Cash versus exchange figures for the April 1934 issue are in the

Annual Report for 1934 (p. 9). For the June sale figures are in New York Times , June 9 1934, "Treasury

Allots Bond and Note Takings."

Appendix 2 Model extension Recall that in the 1930s a zero return to overnight lending meant quoted rates for call money and fed funds were at positive lower bounds, with no demand for loans at these rates. To reproduce this, add to the model a transactions cost paid by overnight lenders that creates a wedge between the return to overnight lending i and the rate quoted to a potential borrower. There is no demand for overnight loans and i = 0 if

33 the quoted rate equals the transactions cost, and the transactions cost exceeds the expected return to holding bond portfolios overnight. Assume this is the case.

Arbitrageurs maximize expression (1) in their own currency. Preferred-habitat investors hold only assets denominated in their own currency. Normalize to one the current exchange rate, expressed as units of foreign currency per dollar. (1+ et ) is the expected value of tomorrow’s exchange rate. et is the expected foreign-currency return to holding American zero-duration assets. Recall American demand for foreign- currency assets was minimal in the 1930s. To reproduce that, add a fixed transactions cost paid by an arbitrageur every period he holds assets valued in the other currency. Assume a relatively high risk of default (political risk) on foreign assets, so the transactions cost deters Americans from holding foreign- currency assets but does not deter foreigners from holding American assets.

The nature of foreign demand for American assets depends on perceived covariances between the exchange rate and local-currency bond values. Here I rely on a few relationships that hold generally. (They hold unless there is negative perceived covariance between day-to-day changes in the exchange rate and changes in the foreign-currency return to holding American bonds.) Foreign arbitrageurs always hold a

f portfolio of American bonds. Its dollar value is N . Its expected overnight dollar return is rft= φ tr t where

rt is the expected return to Americans’ bond portfolio and φ reflects the possibly-different allocation of investments in foreigners’ portfolio. Foreigners may or may not hold American zero-duration assets in a

* period, depending on the value of et relative to the return to foreign-currency overnight lending it and the degree of exchange-rate uncertainty (among other things). If et is relatively large foreigners hold

f f f American zero-duration assets Zt > 0 (in dollars) such that ∂Zt/ ∂ r t=φ t ∂ Z t / ∂ r ft < 0 and

ff fff ∂+(NZ )/ttt ∂=∂r φ ( N+ Z )/ tt ∂ r ≈ 0 . That is, a change in the expected overnight return to American bonds causes foreign arbitrageurs to shift between American zero-duration assets and American bonds, with approximately no change in total dollar value of their American assets. But if e is relatively small

f f f f foreigners hold no American zero-duration assets ( Z t = 0 ) and ∂Nt/ ∂=∂r tφ t N t / ∂> r t 0 , that is an 34 increase in the expected return to American bonds raises foreign demand for American assets, because:

f * *  * 1 Ntttftt=αα1 2 ()(1 ++−+− r )(1 e ) (1 i t ) α3 ttt rWwhere α 1 t ≈ 2   a(1 −ω ) (7) t+1 1 ωt+1 α2t ≈2 2 2 α 3 t ≈ * * σft(1+e t) (1 + r ft ) σ ft σ * t(1 + e t )( 1+r t )(1 + r t + i t )

f f * * * From (7), ∂Nt/ ∂ r t > 0 . W is foreign arbitrageurs’ wealth. rt is the spread between it and the expected

2 foreign-currency return to foreign arbitraguers’ portfolio of foreign bonds. σ *t+ 1 is the variance of the log

2 foreign-currency value of the foreign-bond portfolio. σ ft +1 is the variance of the log value of foreigners'

American bond portfolio in foreign currency (incorporating beliefs about relationships between American

bond prices and the exchange rate). ωt+1 is the correlation between the log foreign-currency values of foreign arbitrageurs’ foreign and American bond portfolios.

Equating supply and demand for American zero-duration assets gives:

S S ph f  rt M tttt+ B − Z − Z (8) 2 = aηt 1−  (1+ rt )  Wt 

f f S S Recall Zt = 0 when (7) holds, otherwise ∂Zt/ ∂ r t < 0 . Thus generally ∂rt/ ∂M t =∂r tt/ ∂ B < 0 .

The change in high-powered money (equivalent to nonborrowed base) across periods is

∆Mt ≈∆ G t −∆ Tres t where G is the value of the Treasury’s gold holdings, Tres is Treasury Money. The

f f change in Gold is ∆≈∆Gt ( NZ+ ) t + X t where X represents exogenous factors affecting the balance of payments such as net exports. Thus, using (8):

ε ε  (9) ∆≈rttα4 [−(G/ M) ∆+∆ g t ( Tres/ M ) tres ttt] ++1 2 where 0 < α 4 ttt ≤ aMW η ( / )

where g (tres) is the log of G (Tres) . Across periods when (7) holds α4 is at the upper bound. For bond k :

β β kt kt ε ε ε (10) ∆ikt ≈α4 t[]−(/) GMgTres ∆ t+ ( /)M ∆ tr es t+ ()1 t + 2 t + 3 kt dk / ζ dk / ζ

ε 1 accounts for effects of changes in the supply of American zero-duration bonds and factors affecting

35 American zero-duration asset demand (such as American arbitrageurs' wealth and preferred-habitat ε investors’ demand). 2 accounts for factors affecting foreign demand for American zero-duration assets

ε (such as e and i* ). 3 accounts for changes in expectations of future overnight rates and/or future values of r (overnight returns to bondholding while the return to overnight lending remains zero).

Now suppose one regresses ∆ikt on (Tres / M ) ∆ tres and (G / M ) ∆ g . Estimated coefficients depend

ε ε ε on correlations between these variables and the omitted variables 1 , 2 , 3 within the sample. In the

ε ε absence of yield-stabilizing operations ∆tres is uncorrelated with 1 and 2 . Because effects of Treasury

ε Money on high-powered money were transitory, they should be uncorrelated with 3 . Thus, in the absence of yield-stabilizing operations the estimated coefficient on (Tres / M ) ∆ tres should be positive. Its

magnitude should increase with βk/ d k . Yield-stabilizing operations create negative correlations between

∆tres and the omitted variables. Operations by the Federal Reserve create negative correlation between ε ∆tres and 1 : as a positive value of ∆tres tends to raise Treasury yields the Federal Reserve would respond

S ε by buying bonds, selling bills; that increases B , decreases 1 . Operations by the Treasury create negative

ε ε ε correlations between ∆tres and all three omitted variables: as an increase in 1 , 2 or 3 raises yields, the

Treasury would buy bonds in response, reducing ∆tres . Thus, the estimated coefficient on

(Tres / M ) ∆ tres is biassed toward zero in a sample with yield-stabilizing operations.

(G / M ) ∆ g may be correlated with the omitted variables with or without yield-stabilizing operations. Across periods when (7) holds:

1 ε ε  ε * (11) ∆≈∆+≈gttttα54 r α 4 t()(/) TresMtres ∆++ ttt 14  whereα 5 t = (/) WG tttt α 12 α φ 1+α4t α 5 t

ε 4 accounts for the exogenous factors that affect foreign demand for American assets ( e, i* , r* ). When (7) does not hold, ∆g is approximately determined only by those exogenous factors. (11) shows that ∆g may

ε be positively correlated with 1 in (10) biassing the coefficient on (G / M ) ∆ g toward a positive value . But

36 ε if variation in ∆g is predominantly due to 4 the coefficient on (G / M ) ∆ g is negative. Because effects of

ε gold purchases on high-powered money were persistent, ∆g could be negatively correlated with 3 (large gold inflow is associated with decreases in expected future overnight rates and/or future r ). Thus the absolute magnitude of the negative coefficient on (G / M ) ∆ g can exceed the coefficient on

(Tres / M ) ∆ tres .

References Ando, Albert and E. Cary Brown. “Lags in Fiscal Policy.” In Stabilization Policies , Commission on Money and Credit, Englewood Cliffs, New Jersey: Prentice-Hall, 1963: 97-163.

Balke, Nathan and Robert J. Gordon. "Appendix: Historical Data" in Robert J. Gordon, ed., The American Business Cycle: Continuity and Change , Chicago: NBER, 1986.

Basile, Peter F., John Landon-Lane and Hugh Rockoff. "Money and Interest Rates in the United States during the Great Depression" in Geoffrey Wood, Terence Mills and Nicholas Crafts, eds., Monetary and Banking History: Essays in honour of Forrest Capie . London: Routledge, 2011: 135-56.

Bauer, Michael D. and Glenn D. Rudebusch. “The Signalling Channel for Federal Reserve Bond Purchases.” International Journal of Central Banking , September 2014, 10(3): 233-289.

Baumeister, Christiane and Luca Benati. "Unconventional Monetary Policy and the Great Recession: Estimating the Macroeconomic Effects of a Spread Compression at the Zero Lower Bound." International Journal of Central Banking, June 2013, 9(2): 165-212.

Beckhart, Benjamin Haggott and James G. Smith. The New York Money Market, Volume II. New York: Columbia Unievrsity Press 1932.

Bernanke, Ben S. “The Macroeconomics of the Great Depression: A Comparative Approach.” Journal of Money, Credit and Banking , February 1995, 27(1): 1-28.

Bernanke, Ben S., Vincent R. Reinhart and Brian P. Sack, “Monetary Policy Alternatives at the Zero Bound: An Empirical Assessment.” Brookings Papers on Economic Activity , 2004:2, 1-78.

Bloomfield, Arthur I. "Operations of the American Exchange Stabilization Fund." Review of Economics and Statistics , May 1944, 26(2): 69-87.

-- Capital Imports and the American Balance of Payments, 1934-39. Chicago: University of Chicago Press, 1950.

Blum, John Morton. From the Morgenthau Diaries: Years of Crisis, 1928-1938. Boston: Houghton-Mifflin, 1959. 37

Bordo, Michael and Arunima Sinha. "A Lesson from the Great Depression that the Federal Reserve Might have Learned." NBER WP 22581, August 2016

Bradford, Frederick A. Money and Banking . New York: Longman, Greens 1941.

Brunner, Karl and Allan H. Meltzer. An Alternative Approach to the Monetary Mechanism. Report to the Subcommittee on Domestic Finance, Committee on Banking and Currency, U.S. House of Representatives. Washington: GPO 1964.

Burgess, Warren Randolph . The Reserve Banks and the Money Market, 2nd edition. New York: Harper and Brothers, 1936.

Carlson, John B. "The Monetary Base, the Economy, and Monetary Policy." Federal Reserve Bank of Cleveland Economic Review , Spring 1981:2-13.

Carter, Susan, Scott Sigmund Gartner, Michael R. Haines, Alan L. Olmstead, Richard Sutch and Gavin Wright. Historical Statistics of the United States, Millenial Edition . Cambridge, UK: Cambridge University Press 2006.

Cecchetti, Stephen G. “The Case of the Negative Nominal Interest Rates: New Estimates of the Term Structure of Interest Rates during the Great Depression.” Journal of Political Economy 96, 1988, 1111- 1141.

Christiano, Lawrence J, Martin Eichenbaum and Charles Evans. “Nominal Rigidities and the Dynamic Effects of a Shock to Monetary Policy.” Journal of Political Economy , February 2005, 113: 1-45.

Chung, Hess, Jean-Phillippe LaForte, David Reifschneider and John C. Williams. "Have We Underestimated the Likelihood and Severity of Zero Lower Bound Events?" Journal of Money, Credit and Banking , Febriuary 2012, 44(1), 47-82.

Clouse, James, Dale Henderson, Athanasios Orphanides, David Small and Peter Tinsley. "Monetary Policy When the Nominal Short-term Interest Rate is Zero." B.E. Journal of Macroeconomics , September 2003, 3(1):1-65.

Daglish, Toby and Lyndon Moore. "U.S. Bond Markets and Credit Spreads during the Great Depression." Working paper, September 2013.

D’Amico, Stefania, William English, David Lopez-Salido and Edward Nelson. “The Federal Reserve’s Large-Scale Asset Purchase Programs: Rationale and Effects.” Economic Journal , November 2012, 122(564): F415-46.

Eggertsson, Gauti B. “Great Expectations and the End of the Depression.” American Economic Review , September 2008, 98(4): 1476-1516.

38 Eggertsson, Gauti B. and Benjamin Pugsley. “The Mistake of 1937: A General Equilibrium Analysis.” Monetary and Economic Studies , December 2006, 24(S-1): 1-40.

Eggertsson, Gauti B. and Michael Woodford. "The Zero Bound on Interest Rates and Optimal Monetary Policy." Brookings Papers on Economic Activity , 2003:1: 139-211.

Ennis, Huberto M/ and Todd Keister. "Understanding Monetary Policy Implementation." Federal Reserve Bank of Richmond Economic Quarterly , Summer 2008, 94(3): 235-263.

Federal Reserve Board of Governors, Annual Report . Washington: GPO, 1934. 1936, 1937, 1938

-- The Federal Reserve System: Its Purposes and Functions. Washington: GPO, 1939.

- Banking and Monetary Statistics , Washington: GPO, 1943.

Friedman, Milton and Anna Jacobson Schwartz. A Monetary History of the United States, 1867-1960. Princeton: Princeton University Press for NBER, 1963.

Fuhrer, Jeffrey C. and Giovanni P. Olivei. "The Estimated Macroeconomic Effects of the Federal Reserve's Large-Scale Treasury Purchase Program." Federal Reserve Bank of Boston Public Policy Brief 11-2, April 2011.

Gagnon, Joseph, Matthew Raskin, Julie Remache, and Brian Sack. "The Financial Market Effects of the Federal Reserve's Large-Scale Asset Purchases." International Journal of Central Banking , March 2011, 7(1): 3-51).

Gagnon, Joseph and Brian Sack. "Monetary Policy with Abundant Liquidity: A New Operating Framework for the Federal Reserve." Peterson Institute for International Economics PB14-4, January 2014.

Goldenweiser, E.A. American Monetary Policy . New York: McGraw-Hill 1951.

Goldsmith, Raymond G. A Study of Saving in the United States, Volume III. Princeton: Princeton University Press, 1956.

-- Financial Intermediaries in the American Economy Since 1900. Princeton: Princeton University Press for NBER, 1958.

Gordon, Robert J. and John Veitch. “Fixed Investment in the American Business Cycle, 1919-1983.” In Robert J. Gordon ed. The American Business Cycle: Continuity and Change. Chicago: University of Chicago Press for NBER, 1986.

Greenlaw, David, James D. Hamilton, Ethan S. Harris and Kenneth D. West. “A Skeptical View of the Impact of the Federal Reserve’s Balance Sheet.” Paper for 2018 U.S. Monetary Policy Forum, University of Chicago, February 2018.

39 Greenwood, Robin and Dimitri Vayanos. "Bond Supply and Excess Bond Returns." Review of Financial Studies , March 2014, 27(3): 663-713.

Griffiss, Bartow. The New York Call Money Market. Johns Hopkins University Dissertation, June 1923.

Hamilton, James D. “Measuring the Liquidity Effect.” American Economic Review , March 1997, 87(1), 80- 97.

Hamilton, James D. and Jing Cynthia Wu. “The Effectiveness of Alternative Monetary Policy Tools in a Zero Lower Bound Environment.” Journal of Money, Credit and Banking , February 2012, 44 supplement (1): 3-46.

Hanes, Christopher. “The Liquidity Trap and U.S. Interest Rates in the 1930s.” Journal of Money, Credit and Banking , February 2006, 38(1): 163-194.

Haney, Lewis, Lyman Logan and Henry S. Gavens. Broker’s Loans . New York: Harper and Brother, 1932.

Hickman, W. Braddock. The Volume of Corporate Bond Financing Since 1900 . Princeton: Princeton University Press for NBER, 1953.

-- Statistical Measures of Corporate Bond Financing Since 1900 . Princeton: Princeton University Press for NBER, 1960.

Hsieh, Chang-Tai aand Christina D. Romer. “Was the Federal Reserve Constrained by the Gold Standard During the Great Depression? Evidence from the 1923 Open Market Purchase Program.” Journal of Economic History , March 2006, 66(1):140-176.

Humpage, Owen and Sanchita Mukherjee. “Even Keel and the Great Inflation.” FRB of Cleveland Working Paper No. 1532, October 2013.

Irwin, Douglas A. “Gold Sterilization and the Recession of 1937-38.” Financial History Review , December 2012, 19(3), 249-267.

Johnson, G. Griffith. The Treasury and Monetary Policy 1933-1938. Cambridge: Harvard University Press, 1939.

Joyce, Michael, David Miles, Andrew Scott and Dimitri Vayanos. "Quantitative Easing and Unconventional Monetary Policy - An Introduction." Economic Journal , Novermber 2012, 122: F271-288,

Krishnamurthy, Arvind and Annette Vissing-Jorgensen. “The Effects of Quantitative Easing on Interest Rates: Channels and Implications for Policy” and “Comments and Discussion.” Brookings Papers on Economic Activity , 2011:1: 215-287.

-- "The Aggregate Demand for Treasury Debt." Journal of Political Economy, April 2012, 120(2): 233-267.

40 Krugman, Paul. "It's Baaack: Japan's Slump and the Return of the Liquidity Trap." Brookings Papers on Economic Activity 1998:2:137-187.

Meltzer, Allan H. "Monetarism: the Issues and the Outcome." Atlantic Economic Journal , March 1998, 26(1): 8-31.

-- A History of the Federal Reserve . Volume I, 1913-1951. Chicago: University of Chicago Press, 2003.

-- A History of the Federal Reserve. Volume II, 1951-1969 . Chicago: University of Chicago Press, 2009.

Modigliani, Franco and Richard Sutch. "Innovations in Interest Rate Policy." American Economic Review , March 1966, 56(1/2): 178-197.

Moody's Investors Service. Corporate Default and Recovery Rates 1920-2010 . February 29, 2011.

Nelson, Edward. "The Future of Monetary Aggregates in Monetary Policy Analysis." Journal of Monetary Economics , March 2003, 50(2): 1029-1059.

Poole, William. "Commercial Bank Reserve management in a Stochastic Model: Implications for Monetary Policy." Journal of Finance , 1968, 23:769-791.

Romer, Christina D. “What Ended the Great Depression?” Journal of Economic History , December 1992, 52(4): 757-784. Sharpe, William F. "Capital Asset prices: A Theory of Market Equilibrium Under Conditions of Risk." Journal of Finance , September 1964, 19:425-42.

Simmons, Edward C. "Treasury Deposits and Excess Reserves." Journal of Political Economy , June 1940, 48(3): 325-343.

Svensson, Lars. "Comments on Bernanke, Reinhart and Sack." Brookings Papers on Economic Activity , 2004:2: 84-93.

Swanson, Eric T. “Let’s Twist Again: A High-frequency Event-Study Analysis of Operation Twist and Its Implications for QE2.” Brookings Papers on Economic Activity , 2011:1: 151-188.

Temin, Peter. Did Monetary Forces Cause the Great Depression? New York: W.W. Norton, 1976.

Tobin, James. “Liquidity Preference as Behavior Toward Risk.” Review of Economic Studies , February 1958, 25: 65-86.

--“The Monetary Interpretation of History.” American Economic Review , June 1965, 55(3): 464-485.

-- "A General Equilibrium Approach to Monetary Theory." Journal of Money, Credit and Banking, February 1969, 1(1): 15-29.

41 Turner, Bernice C. The Federal Fund Market . New York: Prentice-Hall, 1931.

Turner, Robert C. Member Bank Borrowing. Columbus, Ohio: Ohio State University Press, 1938.

U.S. Treasury, Annual Report of the Secretary of the Treasury . Washington: GPO, 1934, 1935, 1936.

Vayanos, Dimitri and Jean-Luc Vila. “A Preferred-Habitat Model of the Term Structure of Interest Rates.” NBER Working Paper 15487, November 2009.

Westerfield, Ray B. Money, Credit and Banking. New York: Ronald Press, 1938.

Wilkins, Mira. The History of Foreign Investment in the United States, 1914-1945 . Cambridge: Harvard University Press, 2004.

Williams, John C. “Lessons from the for Unconventional Monetary Policy.” Remarks at NBER conference, October 2013.

Willis, Parker B. The Federal Funds Market: Its Origin and Development . Boston: Federal Reserve Bank of Boston, 1957.

Woodford, Michael. “Methods of Policy Accommodation at the Interest-Rate Lower Bound.” In Federal Reserve Bank of Kansas City, The Changing Policy Landscape: A Symposium Sponsored By the Federal Reserve Bank of Kansas City At Jackson Hole, Wyoming Aug. 30th - Sept. 12, 2012. Kansas City: Federal Reserve Bank of Kansas City, 2013.

Wright, Jonathan H. “What Does Monetary Policy Do to Long-term Interest Rates at the Zero Lower Bound?” Economic Journal , November 2012, 122(564): F447-66.

42 Tables

Table 1 Federal Reserve Holdings of Long-maturity Treasury Debt and Yields, January 1937-July 1939 Coefficient [Robust (White) SE] p-value LHS variable: Week-to-week change in Federal Reserve Holdings (face value) of Treasury Debt > 5 years' Maturity, $Millions January 1937-July 1939 1 April 1938-July 1939 2 (1) (2) (3) ∆ Treasury bonds 239.37 yield [92.87] 0.01 ∆ Treasury notes 148.34 yield [71.81] 0.04 ∆ Treasury Money 0.046 [0.013] 0.00

∆ Gold 0.015 [0.027] 0.59

N 133 133 66 R2 0.14 0.13 0.18 1Week ending 1/8/ 1937 through week ending 7/29/1939 2Week ending 4/23/1938 through week ending 7/29/1939

Table 2 Characterisitics of Treasury Debt excluding bills Average maturity (years) 1 Percent maturity > 5 years 2 Ratio to GNP 3 Date (1) (2) (3) 1934 June 9.25 50.4 0.29 1935 June 10.03 52.8 0.30 1936 June 9.71 60.1 0.31 December 10.57 62.6 0.30 1937 June 10.44 62.5 0.29 December 10.12 63.3 0.32 1938 June 10.58 64.9 0.33 December 10.95 68.5 0.32 1939 June 11.51 72.4 0.33 December 11.67 76.0 0.29 1 Debt outstanding: held by public, Federal Reserve or Treasury controlled agencies. 2 Debt outstanding less Federal Reserve holdings: held by public or Treasury-controlled agencies 3 Debt held by public/Nominal GNP Sources: Debt outstanding and maturities of marketable public issues from Federal Reserve Board (1943) Table 147. To estimate average maturity I use category midpoints, 25 years for "over 20." Holdings of Federral Reserve banks June 1936-1939 from Federal Reserve Board (1943) Table 91 (I use July 1936 for June 1936). Federal Reserve holdings 1934-35 from Harrison Papers, Box 36, Binder 88, "Exhibit A for FOMC Meeting December 17, 1935" and "Exhibit A for FOMC Executive Committee Meeting February 26, 1936" (assuming maturity distribution same as for the System Account); Treasury-controlled agency holdings (Direct obligations, market issues) from Treasury Bulletin (February 1944:45; January 1939:11; April 1940:24). Nominal GNP from Balke and Gordon (1986: Table 2).

Table 3 Weekly changes in Nonborrowed base, Nonborrowed base plus Treasury Bills, Treasury Money and Gold Coefficient [SE] p-value A) LHS variable: ∆M B) LHS variable: ∆(M + Bills ) Before During After Before During After sterilization 1 sterilization 2 sterilization 3 sterilization 1 sterilization 2 sterilization 3 ∆TresMoney -0.96 -0.99 -0.98 -1.03 -0.63 -0.97 [0.01] [0.03] [0.02] [0.03] [0.18] [0.04] 0.00 0.00 0.00 0.00 0.00 0.00

∆Gold 1.02 1.03 1.01 1.35 0.92 1.20 [0.04] [0.09] [0.03] [0.13] [0.42] [0.13] 0.00 0.00 0.00 0.00 0.03 0.00

N 145 70 66 145 70 66 R2 0.99 0.91 0.99 0.92 0.18 0.91 1Week ending 3/3/1934 through week ending 12/12/1936 2 Week ending 12/19/36 through week ending 4/16/38 3 Week ending 4/30/1938 through week ending 7/29/1939

Table 4 Serial correlation coefficients, factors affecting nonborrowed base

Before sterilization March 3, 1934-December 12, 1936

Lag One week Two weeks Three weeks ∆ Gold 0.63 0.39 0.17 ∆ Treasury Money -0.04 -0.09 -0.13

During sterilization December 19,1936-April 16, 1938

Lag One week Two weeks Three weeks ∆ Gold 0.63 0.60 0.56 ∆ Treasury Money 0.11 -0.04 -0.04

After sterilization April 30, 1938-July 30, 1939

Lag One week Two weeks Three weeks ∆ Gold 0.77 0.60 0.40 ∆ Treasury Money -0.03 -0.09 -0.05

44

Table 5 Bond yields and factors affecting nonborrowed base

Coefficient [Robust (White) Standard error] p-value

LHS variable: weekly change in bond yield A) March 1934-July 1936 1 B)Late April 1938-July 1939 2 C) Late Sept. 1937-July 1939 Treasuries Treasuries Treasuries 3-5 year Long term BAA 3-5 year Long term BAA 3-5 year Long term BAA (Tres / M ) ∆ tres 1.044 0.211 0.812 0.197 -0.029 0.618 0.283 -0.019 1.064 [0.334] [0.148] [0.343] [0.490] [0.368] [0.823] [0.490] [0.342] [1.034] 0.00 0.16 0.02 0.69 0.94 0.46 0.58 0.96 0.31

(G / M ) ∆ g -6.210 -1.633 -4.001 1.097 0.345 9.594 0.827 0.160 7.887 [1.961] [0.787] [1.946] [3.356] [1.942] [4.327] [2.942] [1.689] [4.42] 0.00 0.04 0.04 0.75 0.86 0.03 0.78 0.92 0.08

Time 0.014 0.005 -0.007 -0.007 -0.007 -0.037 0.006 0.005 -0.064 [0.008] [0.003] [0.008] [0.029] [0.020] [0.091] [0.010] [0.006] [0.034] 0.08 0.08 0.38 0.82 0.74 0.69 0.52 0.92 0.06

Time 2 /1000 -0.001 -0.003 7.496 0.003 0.000 -0.004 -0.003 -0.003 0.308 [0.001] [0.002] [8.430] [0.014] [0.001] [0.091] [0.005] [0.003] [0.165] 0.09 0.08 0.38 0.82 0.75 0.69 0.53 0.88 0.06

N 123 123 123 65 65 65 93 93 93 R2 0.16 0.07 0.06 0.01 0.02 0.08 0.02 0.02 0.09

1Week ending 3/3/1934 through week ending 7/11/1936 2 Week ending 4/30/1938 through week ending 7/29/1939 3 Week ending 9/25/1937 through 4/9/1938 and week ending 4/30/1938 through week ending 7/29/1939 Table 6 Correlations between Weekly Debt Sales, Changes in Treasury Money and Treasury Deposits at Commercial Banks, March 1934-July 1936 1 Bond sales Note sales Exchange sales Cash Total Cash Total Bonds Notes ∆ Treasury Money 0.76 0.69 0.68 0.52 0.48 0.23 ∆ Treasury bank balances 0.67 0.47 0.82 0.69 0.24 0.36

Table 7 Bond yields, factors affecting nonborrowed base and debt sales, March 1934-July 1936 1 N = 123 Coefficient [Robust (White) Standard error] p-value A) Treasury 3-5 year yields BAA yields Bond sales Note sales Exchange sales Bond sales Note sales Exchange sales Cash Total Cash Total Bonds Notes Cash Total Cash Total Bonds Notes $Debt sales 83.845 49.758 95.407 50.405 63.238 47.768 34.728 9.529 30.578 10.058 5.009 3.294 ($bn/1000) [32.091] [19.901] [45.290] [30.299] [39.814] [35.113 [22.176] [13.006] [23.644] [19.981] [19.871] [35.131] 0.01 0.01 0.04 0.10 0.11 0.18 0.12 0.47 0.20 0.62 0.80 0.93

(G / M ) ∆ g -6.176 -6.087 -5.953 -6.036 -6.081 -6.175 -4.276 -4.267 -4.208 -4.256 -4.280 -4.289 [2.483] [2.489] [2.520] [2.481] [2.497] [2.476] [1.985] [1.986] [1.992] [2.009] [1.986] [1.993] 0.01 0.02 0.02 0.02 0.02 0.01 0.03 0.03 0.04 0.04 0.03 0.03

R2 0.08 0.09 0.09 0.09 0.09 0.07 0.03 0.03 0.03 0.03 0.03 0.03

B) Treasury 3-5 year yields BAA yields Bond sales Note sales Exchange sales Bond sales Note sales Exchange sales Cash Total Cash Total Bonds Notes Cash Total Cash Total Bonds Notes Debt sales -24.670 19.732 25.109 23.224 32.470 27.215 -66.128 -31.093 -46.148 -18.085 -27.257 -12.755 ($bn/ 1000) [68.837] [38.360] [70.604] [34.102] [48.156] [34.383] [57.395] [25.187] [44.140] [25.531] [27.383] [36.205] 0.72 0.61 0.72 0.50 0.50 0.43 0.25 0.22 0.30 0.48 0.32 0.73

(Tres / M ) ∆ tres 1.285 0.872 0.973 0.911 0.912 1.043 1.194 1.179 1.061 0.944 0.956 0.815 [0.750] [0.707] [0.596] [0.455] [0.500] [0.346] [0.619] [0.544] [0.558] [0.464] [0.394] [0.358] 0.09 0.22 0.11 0.05 0.07 0.00 0.06 0.03 0.06 0.04 0.02 0.02

(G / M ) ∆ g -5.919 -5.956 -5.914 -5.915 -5.929 -5.944 -4.037] -4.090 -4.165 -4.131 -4.121 -4.108 [2.248] [2.328] [2.338] [2.319] [2.328] [2.283] 2.046] [2.047] [2.0600] [2.027] [2.037] [2.019] 0.00 0.01 0.01 0.01 0.01 0.01 0.05 0.05 0.04 0.04 0.05 0.04

R2 0.11 0.11 0.11 0.11 0.11 0.11 0.06 0.06 0.05 0.05 0.06 0.05 1Week ending 3/3/1934 through week ending 7/11/1936 46 Table 8 Yields and Changes in Treasury Money versus Treasury Commercial Bank Deposits (both $Millions) March 1934-July 1936 1

LHS variable: weekly change in bond yield Coefficient [Robust (White) Standard error] p-value

Treasury 3-5 year yields BAA yields ∆Tres /10,000 0.874 0.818 0.642 0.730 [0.258] [0.331] [0.277] [0.346] 0.00 0.01 0.02 0.04

∆TresComm /10,000 0.450 0.0024 0.144 -0.236 [0.452] [0.495] [0.278] [0.284] 0.32 0.96 0.60 0.41

∆g( G / M ) -6.319 -4.427 -4.555 -4.092 -4.331 -4.445 [1.996] [2.208] [2.138] [1.938] [2.165] [2.151] 0.00 0.05 0.04 0.04 0.05 0.04

Time 0.014 0.008 0.008 -0.007 -0.008 -0.008 [0.008] [0.008] [0.008] [0.008] [0.009] [0.009] 0.07 0.30 0.33 0.40 0.37 0.34

Time 2 /1000 -0.008 -0.005 -0.004 0.004 0.005 0.005 [0.004] [0.005] [0.005] [0.005] [0.005] [0.005] 0.08 0.31 0.34 0.41 0.37 0.35

R2 0.16 0.06 0.09 0.06 0.04 0.06

1Week ending 3/3/1934 through week ending 7/11/1936

47 Figure 1 Interest Rates February 1934-December 1939 Weekly (except Treasury 10-year) P 7 1 a 2 3 a b 4 5 6 T

6 BAA

5

4 t Treasury n

e Long-term bonds c r

e 10-year p 3

2 Treasury 3-5 year

1

Treasury bills 9-month auction 3-month dealer bid 0 34 35 36 37 38 39 Sources: Federal Reserve Board (1943) Tables 123, 129; Cecchetti (1988).

Figure 2 Nonborrowed base (SA) and Treasury Bills February 1934-December 1939 Monthly (end of month) P T 22,000 11a 2 3 3a 3b 4 5 6

20,000

18,000

16,000

Nonborrowed base plus Treasury Bills

$millioins 14,000

12,000

10,000 Nonborrowed base

8,000 Jan 35 Jan 36 Jan 37 Jan 38 Jan 39 Sources: Nonborrowed base is is Friedman and Schwartz (1963), Table B-3, column (1), minus Federal Reserve discount lending (last week in month) from Federal Reserve Board (1943), Table 103. Treasury bills held by the public is bills outstanding, Federal Reserve Board (1943), Table 146, "Treasury bills," minus Federal Reserve bill holdings, Federal Reserve Board (1943), Table 91, "Bills." Treasury-controlled agencies do not appear to have held bills (Treasury Bulletin January 1939:16; and subsequent issues).

49 Figure 3 10-year Treasury Yield and Nonborrowed base March 1934-July 1939 Monthly (end of month)

3.6

3.2

2.8

2.4 percent

2.0

1.6

1.2 9.1 9.2 9.3 9.4 9.5 9.6 9.7 9.8

Log nonborrowed base

50 Figure 4 10-year Treasury Yield and Nonborrowed base plus Bills March 1934-July 1939 Monthly (end of month) 3.6

3.2

2.8

2.4 percent

2.0

April 1934-November 1936 1.6 December 1936-April 1938 May 1938-July 1939

1.2 9.2 9.3 9.4 9.5 9.6 9.7 9.8 9.9

Log nonborrowed base plus bills

51 Figure 5 Nonborrowed base and Components February 1934-July 1939 Weekly, Wednesday

1 20,000 a 2 3 a b 4 5 6

16,000

12,000 Nonborrowed base

Gold $Millions 8,000

4,000 Treasury Money

Fed Securities Portfolio 0 Other 34 35 36 37 38 39 Week ending 2/15/1934 through week ending 7/31/1939

Source: Federal Reserve Board (1943), Table 103. Treasury Money is sum of "Treasury cash holdings" and "Treasury deposits with Federal Reserve Banks." Gold is "Gold stock." Federal Reserve Securities portfolio is "Federal Reserve Bank credit outstanding/U.S. Government securities." Other is "Treasury currency outstanding" (includes effects of Treasury silver purchases). Nonborrowed base is "Member Bank Reserve Balances" plus "Nonmember Deposits" plus "Money in Circulation" minus "Federal Reserve Bank credit outstanding/Bills discounted." Estimate for nonborrowed base rescaled so that value for last week in a month matches the value for Figure 2.

52