July 7, 2010

Economics Research Ernest W. (Chip) Brown, Head 212-583-4663 [email protected] Who Wants to Be a Millionaire? Alexandre Schwartsman 5511-3012-5726 [email protected] • Those who follow the evolution of fiscal accounts in Brazil closely surely have noticed something about the interest payments on the public sector debt. After a significant (and more or less steady) decline from about 9.5% of GDP in 3Q03 to little less than 5% of GDP in 3Q09, interest payments have started to increase again, standing at 5.4% of GDP during the 12-month period ended in May 2010. • The implicit cost of net domestic federal debt used to follow quite closely the movements of the Selic rate, but the correlation broke down in the past two years: while the policy rate declined, the implicit debt cost has gone up. This development, however, has occurred against a backdrop of lower interest rates and a reduced net debt, which is something of a puzzle. • In order to solve this puzzle we put forward the following hypothesis: the rise in interest payments results from the negative spread between the interest rates received by the government and the it pays on its debt. This spread was less important until 2008, when government assets were smaller, but has become more relevant in the last two years, as government lending to BNDES jumped from 0.5% to more than 6% of GDP. • An alternative explanation could be the change in the debt profile, resulting from a higher share of fixed interest rate and linked securities at the expense of Selic-linked notes. Yet, once we include the Central Bank repos in the picture we cannot find a significant difference between the debt profile in 2004-2007 (when the correlation between the debt cost and the Selic rate was positive and high) and during the 2008-2010 period, when the correlation broke down. • With the help of a simple example we find that the implicit debt cost would be close to the Selic rate if one of the two conditions is valid: (1) the spread between the Selic rate and the return on government is small; or (2) the ratio between government assets and the net debt is small. The first condition is typically not true, but it was the second one that ceased to be valid in 2008-2010 on the back of the fast expansion of National Treasury credits to BNDES, which rose from about 1% of the net debt to 12% of the net debt in that period. • The negative correlation between the Selic rate and the implicit debt cost observed in the past two years is not, however, a structural feature. As the spread between the Selic rate and the rate at which the government lends to BNDES rises, the implicit cost should go up as well, an effect now likely to be magnified by a higher share of BNDES credits relative to net debt. • Our calculations do not take into consideration an additional channel. Given that increased funding leads to BNDES leads to higher lending, it should expand domestic demand, requiring an additional effort in terms of the Selic rate than it would be necessary in the absence of such policy. Hence, the spread between the Selic rate and the rate at which the government lends to BNDES is likely to rise even more, leading to a further increase in the interest bill. Those who follow closely the evolution of fiscal accounts in Brazil surely have noticed something about the interest payments on the public sector debt. After a significant (and more or less steady) decline from about 9.5% of GDP in 3Q03 to little less than 5% of GDP in 3Q09, interest payments have started to increase again, standing at 5.4% of GDP during the 12-month period ended in May 2010. This development is hard to square with the slight decline of the net debt relative to the levels

U.S. investors’ inquiries should be directed to Santander Investment at (212) 350-0707. observed in 3Q09 and the stability of the Selic from 3Q09 until the beginning of 2Q10.1 There must be something odd going on when interest payments rise while the debt has declined and interest rates barely budged.

Interest on debt - % GDP Interest rates & net debt 9.5% 33% 60% Selic (left) 360-day swap (left) 8.5% 28% Net debt (right) 55%

7.5% 23% 50%

6.5% 18% 45% Net debt - % GDP

5.5% Interest rate - % per annum 13% 40%

4.5% 8% 35% 0 0 1 2 2 3 4 5 6 7 8 9 0 -0 0 -0 -01 -0 0 -0 -0 -04 -0 -0 -06 -0 07 -0 -0 -09 -1 1 2 3 4 4 5 5 6 6 6 7 8 9 9 0 0 y y n y n y- y -01 -0 -03 -0 0 -0 -0 -0 -08 -08 -0 0 a a ep-04 a ep-06 a a n p p n n y n p n Jan May-Sep-00Jan M Sep-01Jan May-Sep-02Jan May-03Sep-03Jan M S Ja May-05Sep-05Jan M S Ja M Sep-07Jan May-08Sep-08Jan M Sep-09Jan May-10 a e ay-0 a a a a Jan-00May-00Sep-00J May-01S Jan-0May-02Sep-02Jan-03M Se J May-04Sep-Jan-0May-0Sep-05J Ma Sep Jan-0May-07Sep-07J May-0Se J May-09Sep-Jan-1May-1 Source: Central Bank.

The implicit cost2 of net domestic federal debt used to follow quite closely the movements of the Selic rate; even though there have always been differences between the level of the debt cost and the Selic rate, they tended to move in line, in particular when we concentrate on the net domestic federal debt cost. As shown below, the observed correlation between the 12-month average of the net domestic federal debt cost and the 12-month average of the Selic rate approached 0.9 between 2004 and 2007, becoming, however, significantly negative at -0.6 between 2008 and 2010 (actually the 12-month period ended in May 2010). That is, the decline of domestic interest rates since late 2008 did not result in a correspondent reduction in the debt cost estimates for most of the period after 2007.

Federal debt cost vs Selic (12-month) 33 Net federal debt Net domestic federal debt 28 Selic

23

18

13

8

4 5 6 7 7 8 9 0 -0 -0 -05 -0 -0 -07 -0 -0 -08 -0 -1 -10 y n y y n y ep-04 a a ep a a Jan May-04S Jan M Sep-05 Jan May-06Sep-06 Ja M S Jan M Sep-08 Ja May-09Sep-09 Jan M

1 It is true that the 360-day swap rate did increase from 9% per annum to about 12% per annum in the period, but note that this would have had an impact only in the new prefixed debt issued between 3Q09 and 2Q10, which would hardly explain the rise in interest payments. 2 The implicit cost of government debt is estimated by the Central Bank. Using the detailed information currently available only to the Central Bank, it estimates the implicit debt cost as the ratio of interest payments and the debt. Specifically, in the case at hand, the Central Bank calculates the cost of the net domestic federal debt as the ratio between net interest (interest accrued on the debt deducted interest accrued on the federal government assets) and the net domestic federal debt.

July 7, 2010 2

Domestic federal debt cost x Selic (2004-07) Domestic federal debt cost x Selic (2008-10) 30 15

28 14 26 13 24

22 12 20 11 18 Domestic federal debt cost Domestic federal debt cost 16 10

14 9 11 13 15 17 19 21 23 8 9 10 11 12 13 Selic Selic

Sources: Central Bank and Santander estimates from Central Bank data.

In order to solve this puzzle we put forward the following hypothesis: the rise in interest payments results from the negative spread between the interest rates received by the government and the interest rate it pays on its debt. This spread was less important until 2008, when government assets were smaller, but has become more relevant in the last two years, as government lending to BNDES jumped from 0.5% to more than 6% of GDP. Of course, this pattern could have originated from the change in the government debt structure, namely the increase in the share of government securities whose yield is not directly linked to the Selic rate – the prefixed debt (LTNs and NTN-Fs) and inflation linked debt (NTN-Bs and NTN-Cs) at the expense of Selic linked debt (LFTs). As more debt had been contracted at fixed rates, it would be only natural to observe a lower correlation between the overnight rate and the debt cost. A closer look, however, suggests that this is unlikely to have led to the dramatic change in the correlation observed above. Whereas it is true that the National Treasury has made a substantial effort to reduce the share of Selic-linked securities, a somewhat wider definition of the federal domestic debt, which also includes the repo operations through which the Central Bank regulates the money supply, reveals that the change in the debt structure becomes less impressive once we include the repos in the picture. The share of Selic-linked notes indeed reduced from nearly 50% to 29% of the federal debt in between December 2004 and May 2010, but, during the same period, repos jumped from 5.5% to 18% of the debt. Taken jointly, we still observe a reduction in overnight-linked debt, from 55% to 47%, but not that remarkable. Moreover, looking at the 2004-2007 average versus the 2008-20100 average, which would be the relevant comparison to understand the different performance in these periods, we find a very modest reduction in overnight-linked debt (LFTs and repos), from 49% to 48%, while the share of prefixed and inflation-linked securities rose from 47% to 50%.

Federal debt duration (excluding repos) 30

25 Dec-07 20

15 Months

Dec-03 10

5

0

8 8 9 9 0 1 4 5 6 7 8 9 -9 -9 -9 -02 -03 0 -04 0 -05 -0 -06 -0 -07 -08 -09 -10 n l n l l l l n l n l n n n a u a u u u u a u a u a a a J J J Jul-9Jan-00Jul-0Jan-01Jul-0Jan-02J Jan-03J Jan- J Jan- J J J J J J Jul-0J Jul-0J Sources: Central Bank and Santander estimates from Central Bank data.

In other words, the federal debt duration figures above overstate, to some extent, the actual increase in duration, since they do not take into consideration the large share of debt currently under the form of Central Bank repos.

July 7, 2010 3

Those figures indicate that we have to look somewhere else to understand the correlation change. In the next section we illustrate through a simple example, how the negative spread between interest received in domestic government assets and interest paid on the domestic debt produces an increase in the implicit debt cost.

A simple example Let “i” denote the interest rate paid on (gross) domestic debt (D), and “i+” be the interest received from government assets (A). The total interest bill, “I,” would therefore be given by the difference between interest accrued on the debt and interest accrued on assets:

−= + AiiDI (1) We define the net debt (N) simply as the difference between the gross debt, D, and assets, A: −= ADN (2) Now adding and subtracting iA in equation (1) and using definition (2) we can find: −+= + )( AiiiNI (3)

Dividing both sides of (3) by the net debt (N) we find the following expression for the implicit domestic debt cost ˆ ≡ / NIi : ˆ −+= + NAiiii )/)(( (4) The expression above suggests that the implicit debt cost should fluctuate in line with the gross debt cost provided one of the two following conditions prevails: (a) the interest rate differential (i-i+) is small3; or (b) the ratio between government assets and the net debt is small. The first condition has not been observed for most of the time as the 12-month spread between the Selic rate and the Long-Term Interest Rate (TJLP, the interest rate that accrues on government lending to BNDES) has usually been high, with the possible exception of the most recent period, when it dropped to the neighborhood of 2% per annum.

Selic (-) TJLP (12-month) BNDES credit/Net Debt - % 12 14

10 12

10 8 8 6 6 4 4 2 2

0 0 0 1 2 3 3 4 5 6 7 7 8 9 0 -0 -0 -01 -0 -0 0 -0 -04 -0 -0 06 -0 -07 0 -0 -0 -09 -1 -10 1 1 2 2 3 4 4 5 6 7 7 8 8 8 9 9 9 y y y- y p- y y -0 -01 0 -04 -0 -05 -0 -0 -06 0 0 0 0 -0 -10 a ep-01 a a a e an a a n p n p n p Jan May-00Sep-00Jan M S Jan May-02Sep-02Jan May-Sep-03Jan M Sep-04Jan May-05Sep-05Jan M Sep-06Jan M S J May-08Sep-08Jan M Sep-09Jan M a a ay-05e a ay-10 J May-0Se Jan-May-0Sep-02Jan-03May-0Sep-03Jan May-0Sep J M S J May-06Se Jan-May-0Sep-07Jan-May-0Sep- Jan-May-0Sep Jan M Sources: Central Bank and Santander estimates from Central Bank data.

However, this did not matter much for the net domestic federal implicit debt cost most of the time, because Treasury lending to BNDES represented a very small proportion of the net debt, typically hovering between 0% and 2% until late in 2008, as the chart above at the right reveals. Since then, however, this share jumped to more than 12% of the net debt (measured relative to GDP it jumped from 0.5% to 6%). Thus, as the second condition failed, the close correlation between the Selic rate and the implicit net domestic federal cost broke down. To put it differently, the National Treasury had to finance additional lending to BNDES through the issuance of domestic debt, so at the end of the day the impact on the net debt was zero.4 However, the National Treasury lends to BNDES at a much lower rate (typically the TJLP) than its borrowing cost, paying, therefore, a negative spread, which, as expressed by equation (4), adds to cost of (gross) domestic debt. Moreover, the larger are the loans to BNDES, the higher is the impact of negative impact on the implicit net domestic federal debt.

3 Alternatively, the negative spread (i+-i) is small 4 On this issue, please refer to our earlier report “Bond, Federal Bond,” January 7, 2010.

July 7, 2010 4

Notice further that the impact is bound to increase. As we know, the Selic is scheduled to increase in the coming months, which means that the spread between the Selic and TJLP is going up as well. Even if the Selic and TJLP were to stay at their current levels (10.25% and 6% respectively), the 12-month spread would increase from 2.6% observed in May 2010 to 4%. If we are correct in our forecasts, the Selic rate should reach 13% per annum at the end of the tightening cycle, which would push the spread further to 6.6% per annum, about 2.5 times higher than the May observation. Hence, regardless of any further increase in credits to BNDES, one should expect the implicit domestic federal debt cost to go up.

Conclusion The observed negative correlation may have led to some hope that the impending rise of the Selic rate would not translate into a similar hike of the debt cost, but this is definitely not the case. The positive correlation broke down because credits to BNDES increased at the same time interest rates came down, preventing their decline from translating into lower debt costs. Now the interest rate hike should affect debt costs through two channels, as suggested by equation (4): the direct impact on debt costs and the increased spread weighed by a higher proportion of government assets relative to the net debt. In other words, the negative correlation was not a structural feature of the economy. If one had any hope that the negative correlation implied that BNDES credits would mean a hedge against interest rates hikes, our advice would be to go over the issue again. As argued above, debt costs are about to rise significantly due to the combination of a higher share of government assets relative to the net debt and the increase in the spread between the gross debt cost and the return on government assets.

BNDES impact on implicit debt cost (basis points) BNDES impact on debt cost/GDP (basis points) 50 18 45 16 40 14 35 12 30 10 25 8 20 15 6 10 4 5 2 0 0 1 1 2 4 5 6 7 8 9 9 0 -0 0 -0 -02 -03 -0 -0 -05 -0 -06 -0 07 -0 -08 -0 0 -1 -10 1 1 2 2 3 4 4 5 6 7 7 8 8 8 9 9 9 y y y y y- y y -0 -01 0 -04 -0 -05 -0 -0 -06 0 0 0 0 -0 -10 a a an a a a a ep-08 a n p n p n p Jan May-Sep-01Jan M Sep-02Jan-03M Sep-03Jan May-04Sep-04J M Sep-05Jan M Sep-06Jan M Sep-07Jan M S Jan May-Sep-09Jan M a a ay-05e a ay-10 J May-0Se Jan-May-0Sep-02Jan-03May-0Sep-03Jan May-0Sep J M S J May-06Se Jan-May-0Sep-07Jan-May-0Sep- Jan-May-0Sep Jan M Source: Santander estimates from Central Bank data.

There are, of course, other important government assets, in particular international reserves, that also have impacts on the implicit cost of the net debt (not only the net federal domestic debt), but, as we have argued in previous research, international reserves display two important differences relative to credits to BNDES. First, international reserves are liquid assets, which can, presumably, help finance the debt rollover in a stress scenario. Second, the local valuation of reserves (that is, reserves translated into domestic currency) should typically increase under bad conditions (a decline in commodity prices, or a sudden stop in capital flows), and decrease in favorable conditions; that is, they play a role of a hedge, whereas BNDES credits, denominated in local currency, do not. That said, there is another issue that deserves further exploration, which we intend to go over in future research. In our simple example above we took the gross debt cost (i.e., the Selic rate) as a given, or, better said, as a variable unaffected by increase or reduction of government assets, but in a more realistic setting this is unlikely to be the case. We can imagine at least two channels through which an increase in government funding for BNDES can lead to higher domestic interest rates. The least important one is a direct result of the previous discussion. If higher funding to BNDES leads to a higher debt cost, and hence to a higher debt path, everything else constant, there might be some impact on sovereign spreads, which could affect domestic rates. But this channel would hardly be a relevant one. At the current debt levels and given primary performance, we are really discussing the speed at which the debt would decline, rather than a process that could lead to much higher rates due to increasing risk premium on government debt. We deem as possibly more important the likely effect that additional funding for BNDES would have on lending ability and hence on domestic demand. Notice, first, how increased National Treasury funding to BNDES from late 2008 onwards (from BRL20 billion to BRL130 billion) led to a correspondent increase in the bank’s leading, which came from about BRL190 billion in 3Q08 to BRL300 billion in 2Q10. The second leg of increased funding (additional BRL80 billion, from BRL130

July 7, 2010 5 billion to BRL210 billion) that took place in late April has yet to translate into additional lending, but this is only a matter of time.

BNDES funding - constant R$ million BNDES lending - constant R$ million 250,000 300,000 280,000 200,000 260,000 240,000 150,000 220,000 200,000 100,000 180,000 160,000 50,000 140,000 120,000 0 100,000 1 2 4 5 6 7 8 0 0 0 0 0 0 0 2 2 3 5 5 6 8 8 9 ------01 0 04 n-05 ul -01 l- - l-0 l-0 -04 l- l-0 l-0 -07 l-0 -09 l-0 -10 Jul Jul Jul-03 Jul Jul an-06 Jul J Jul Jul-09 n u n u u n u n-0 u u n n-0 u n u n Jan-01 Jan-02 Jan-03 Jan-04 Ja J Jan-07 Jan-08 Jan-09 Jan-10 a J a J J a J a J J a Jul-07 a J a J a J J Jan-03 J J Jan-06 J J J J Sources: Central Bank.

At the same time, there is little doubt that the significant increase of BNDES balance sheet (BRL110 billion so far, and possibly BRL80 billion more to come) should have had a substantial impact on domestic demand. We have argued5 that this policy shares more than a passing resemblance to fiscal policy and, indeed, many have classified BNDES credit expansion as quasi- fiscal policy. But then, as is usually the case, everything else constant, a more expansionary fiscal policy requires a more contractionary monetary policy (higher interest rates) to deliver the same level of inflation. Extending the reasoning to the increase of BNDES funding (therefore its lending), it is straightforward to conclude that the increase in government assets also implies higher interest rates than the ones that would prevail in the absence of such policy. Thus, the impact of higher government assets is not limited to being a channel through which the negative spread between returns on assets and interest on debt increases the implicit debt cost; it also implies an additional increase of the negative spread through the required rise in the Selic rate to offset the expansionary impact of BNDES lending on demand, magnifying the effect described in the preceding section. It is true that BNDES financing should have some impact on investment, hence on potential GDP growth. But this would hardly be strong enough to offset the impact on demand. For a start, investment is first demand, becoming additional supply only after a while. In a recent paper6 we estimated that investment becomes new industrial capacity after some six quarters, being of little use to help with current inflationary pressures. Second, we also estimate that, in order to boost potential GDP growth by 1% per annum, investment needs to increase between 4% and 5% of GDP. Given that investment is currently 18% of GDP, it would have to reach, conservatively, 22% of GDP to accelerate potential growth from, say, 4.5% to 5.5% per annum, an expansion of 22%. In order to reach this 22% increase in the investment rate in one year, assuming that the economy would grow at potential7 (4.5%), gross capital formation would have to increase a little less than 28% in a single year. To reach this target over two years, investment would have to increase about 16% per annum during those years. In any case, domestic demand would have to accelerate drastically due to investment growth before any additional supply would appear. Note, moreover, that the economy is growing much faster than potential (about 9-10% per annum at the margin), so, for investment to accommodate current growth, it would not be enough to increase investment only by 4% of GDP, but a larger figure instead, a consideration that only makes the problem much harder to solve. In short, the notion that one can offset inflationary pressures through the increase in supply resulting from higher investment does not really stand on solid ground. For investment to have a meaningful impact on supply, it would have to increase massively, propping up current demand. Second, it would not become additional supply immediately, but after about a year and half. In the meantime, the inflationary problem would only get worse, requiring further monetary policy tightening. Thus, going back to the original issue, it seems reasonable to conclude that the increase in the availability of credit at below- market interest rates results in further inflationary pressures, and thus higher policy rates. This would feed back into the spread between the gross debt cost and the return on government assets, amplifying the effect described earlier.

5 “Bond, Federal Bond,” January 7, 2010, page 6. 6 “576 Regressions on Capacity Utilization,” April 26, 2010. 7 If growth is higher than potential, the increase in investment would have to be even higher, so our conclusions do not really depend on this assumption. July 7, 2010 6

We still do not think that such developments would lead to a deterioration of the debt dynamics large enough to cause a debt crisis. However, we had better get used to a considerably higher implicit debt cost, and, therefore, an interest bill permanently higher than the one that would have prevailed in the absence of the expansion of government assets. Summing up, the expansion of BNDES credits implies a higher government transfer for those who have access to cheap financing and can appropriate the negative spread between the TJLP and the Selic rate – a riskless path to becoming a millionaire.

July 7, 2010 7

CONTACTS Economics Research Ernest (Chip) Brown Head of Economics Research [email protected] 212-583-4663 Sergio Galván – Argentina [email protected] 54-11-4341-1728 Alexandre Schwartsman Economist – Brazil [email protected] 5511-3012-5726 Juan Pablo Castro Economist – Chile [email protected] 562-336-3389 Felipe Hernandez Calle Economist – Colombia [email protected] 571-644-8006 Delia Paredes Economist – Mexico [email protected] 5255-5269-1932 Fixed Income Research Fernando Marin Head of Global Research [email protected] 3491-257-2100 Juan Pablo Cabrera Senior Economist – Local Markets [email protected] 3491-257-2172 Alejandro Estevez-Breton Local Markets Strategy [email protected] 212-350-3917 Equity Research Cristián Moreno Head, Equity Research [email protected] 212-350-3992 Walter Chiarvesio Head, Argentina [email protected] 5411-4341-1564 Marcelo Audi Head, Brazil [email protected] 5511-3012-5749 Francisco Errandonea Head, Chile [email protected] 562-336-3357 Gonzalo Fernandez Head, Mexico [email protected] 5255-5269-1931 Electronic Media Bloomberg STDR Reuters Pages SISEMA through SISEMZ

This report has been prepared by Santander Investment Securities Inc. (“SIS”) (a subsidiary of Santander Investment I S.A., which is wholly owned by Banco Santander, S.A. (“Santander”), on behalf of itself and its affiliates (collectively, Grupo Santander) and is provided for information purposes only. This document must not be considered as an offer to sell or a solicitation of an offer to buy any relevant securities (i.e., securities mentioned herein or of the same issuer and/or options, warrants, or rights with respect to or interests in any such securities). Any decision by the recipient to buy or to sell should be based on publicly available information on the related security and, where appropriate, should take into account the content of the related prospectus filed with and available from the entity governing the related market and the company issuing the security. This report is issued in Spain by Santander Investment Bolsa, Sociedad de Valores, S.A. (“Santander Investment Bolsa”), and in the United Kingdom by Banco Santander, S.A., London Branch. Santander London is authorized by the Bank of Spain. This report is not being issued to private customers. SIS, Santander London and Santander Investment Bolsa are members of Grupo Santander. ANALYST CERTIFICATION: The following analysts hereby certify that their views about the companies and their securities discussed in this report are accurately expressed, that their recommendations reflect solely and exclusively their personal opinions, and that such opinions were prepared in an independent and autonomous manner, including as regards the institution to which they are linked, and that they have not received and will not receive direct or indirect compensation in exchange for expressing specific recommendations or views in this report, since their compensation and the compensation system applying to Grupo Santander and any of its affiliates is not pegged to the pricing of any of the securities issued by the companies evaluated in the report, or to the income arising from the businesses and financial transactions carried out by Grupo Santander and any of its affiliates: Ernest W. (Chip) Brown, Alexandre Schwartsman. The information contained herein has been compiled from sources believed to be reliable, but, although all reasonable care has been taken to ensure that the information contained herein is not untrue or misleading, we make no representation that it is accurate or complete and it should not be relied upon as such. All opinions and estimates included herein constitute our judgment as at the date of this report and are subject to change without notice. Any U.S. recipient of this report (other than a registered broker-dealer or a bank acting in a broker-dealer capacity) that would like to effect any transaction in any security discussed herein should contact and place orders in the United States with SIS, which, without in any way limiting the foregoing, accepts responsibility (solely for purposes of and within the meaning of Rule 15a-6 under the U.S. Securities Exchange Act of 1934) for this report and its dissemination in the United States. © 2010 by Santander Investment Securities Inc. All Rights Reserved.

July 7, 2010 8