Financial Intermediaries and International Risk Premia

Transcription

1 Financial Intermediaries and International Risk Premia Kyriakos Chousakos October 2017 Abstract I propose a measure of adjusted leverage as a proxy for the pricing kernel of a representative financial intermediary. Using a simple theoretical framework with information production, I show that in states of the world where credit outstanding in the economy is low, financial leverage is not an accurate proxy for the stochastic discount factor of financial intermediaries. Empirical evidence confirms this theoretical finding for an international panel of financial intermediaries. Credit outstanding arises as an important determinant of the stochastic discount factor of a financial intermediary. As a result, the new proposed measure incorporates information on both intermediaries financial leverage and the amount of credit in the economy. It is an economically meaningful state variable that is pro-cyclical and predicts financial crises. I show that a global adjusted leverage factor prices currency portfolios and global equity portfolios outperforming benchmark factor models designed to price these assets. Keywords: Asset Pricing, International Financial Markets, Financial Intermediaries JEL Classification: G2, G12, G15, G24 Yale University. kyriakos.chousakos@yale.edu. I am grateful to my advisors Gary Gorton, Tobias Moskowitz, and Jonathan Ingersoll for their guidance and invaluable comments. I thank Oliver Boguth (discussant), Robin Greenwood, Tyler Muir (discussant), and Guillermo Ordoñez for helpful comments and suggestions. I also thank Thomas Bonzcek, Arun Gupta, Toomas Laarits, Avner Langut, Adriana Robertson, and participants of the 5th Annual USC Marshall PhD Conference in Finance, the PhD Session of the 2017 Northern Finance Association Annual Conference, and the PanAgora 2017 Crowell Prize Seminar Series for useful comments. All errors remain my own.

2 1 Introduction Financial intermediaries are the primary participants in capital markets. In the foreign exchange market, international commercial banks acting as securities dealers account for more than 51% of all transactions. 1 In the equities market, over the past decades households have been steadily decreasing their direct stock holdings, while financial intermediaries have been filling the void. 2 In the bond market, almost all of the trading takes place between broker-dealers and large institutional investors in over-the-counter markets. 3 Financial intermediaries are sophisticated market participants that carry out complex trading strategies, face low transactions costs, and continuously update their strategies as new information becomes available. As a result, financial intermediaries are ideal candidates for the role of the marginal investor in a wide array of markets which means that their marginal value of wealth is expected to price financial assets in these markets. 4 In this paper, I improve on existing intermediary asset pricing models and study the impact of financial intermediaries on global capital markets. More specifically, shifting the focus from U.S. only to international financial intermediaries, I propose and test an empirical proxy for the pricing kernel of a representative global financial intermediary. This proxy in addition to financial leverage takes into account the amount of credit in the economy. This is motivated by a simple theoretical framework with information production in which credit outstanding in the economy along with financial leverage arises as an important determinant of the stochastic discount factor (SDF) of financial intermediaries. Empirical evidence presented in the paper confirms this theoretical finding. The proposed measure is an adjusted leverage index which incorporates information on intermediaries financial leverage and the availability of credit in the economy. I show that adjusted leverage is an economically meaningful state variable that is pro-cyclical and predicts financial crises and future consumption levels. I find that a global adjusted leverage factor prices currency portfolios and global equity portfolios outperforming benchmark factor models designed to price these assets. A decomposition of the global adjusted leverage factor into non-u.s. and U.S. only components reveals that non-u.s. financial intermediaries are marginal investors in foreign exchange markets as well as in global equity markets. To motivate the empirical work of this paper, I develop a simple theoretical framework in the spirit of Gorton and Ordoñez (2014) and Gorton and Ordoñez (2016). The economy comprises three agents firms, financial institutions, and households all of which are risk neutral with respect to lending activities 1 See, the foreign exchange turnover section of the triennial central bank survey conducted by the Bank for International Settlements (BIS) in September 2016 ( 2 See, e.g. Allen (2001), and Sneider et al. (2013). 3 See, e.g. Edwards et al. (2007) for a breakdown of the percentage of bonds traded over-the-counter and in NYSE. 4 A growing stream of the literature, both theoretical and empirical, studies the relation between financial intermediaries and asset prices. I discuss the different approaches in Section 2. 1

3 with the exception of financial institutions which are risk averse with regard to holding firms equity. 5 To produce output, firms need to borrow capital from households through the financial system posting land as collateral. Both households and financial institutions may produce information regarding the quality of the collateral backing deposits and loans respectively with the cost of producing information for households being significantly higher compared to that for financial institutions. 6 In this setting, information production regulates the amount of credit outstanding and deposits in the economy. Both quantities are an increasing function of expected output and their respective rates of increase depend on the current information production regime. In this framework, I show that the relation between the SDF, as described in the model by the marginal value of wealth of a financial institution, and its financial leverage is not one-to-one and strongly depends on the level of credit outstanding. An increase in financial leverage does not necessarily indicate a decrease in the marginal value of wealth for the financial institution. This is primarily observed in times of economic growth, where credit outstanding is high and financial intermediaries sustain high financial leverage and low marginal value of wealth as a result of the ample investment opportunities which they can undertake. However, this is not the case in times of recessions or financial crises, where the amount of credit outstanding is low. In such times both financial leverage and the marginal value of wealth of a financial institution increase as a result of an increase in deposits not followed by a similar increase in loans and profitability. According to the theoretical framework discussed above, it is possible that in a low credit environment an increase in leverage is associated with an increase in the marginal value of wealth of the financial institution. This implies that financial leverage is not an accurate empirical proxy for the SDF of a financial intermediary. A number of empirical findings, summarized below, corroborate this theoretical proposition. Assets that co-vary with intermediaries SDF are riskier and investors require higher premia to compensate for that risk. In the literature the SDF of financial intermediaries is proxied by leverage innovations of financial intermediaries (see, e.g. Adrian et al. (2014)) or its reciprocal capital ratio innovations (see, e.g. He et al. (forthcoming)). I empirically show that for an international panel of countries financial leverage interacts differently with key characteristics of financial intermediaries, such as future financial assets and stock market returns, depending on the level of credit outstanding in the economy. When the credit outstanding is high, financial leverage is positively correlated with the level of future assets reflecting a higher risk bearing capacity and negatively correlated with market returns indicating lower risk for such investments. On the other hand, 5 This assumption is primarily motivated by the third Basel Accord (Basel III) according to which secure debt is considered to be more liquid than equity and as a result the capital requirements for financial institutions holding equities in their balance sheets are higher compared to these for secure debt assets. 6 Financial institutions possess superior technology and resources in identifying the quality of collateral posted by firms compared to that of households. 2

4 when the credit outstanding in the economy is low the opposite holds true. This finding suggests that a potential proxy for the SDF of financial intermediaries ought to take into account credit outstanding in the economy. I construct a proxy for intermediaries SDF by combining information on financial leverage of brokerdealers with information on economy-wide credit-to-private sector. More specifically, first, I compute a measure of global financial leverage as the aggregated country level financial leverage weighted by the level of financial assets of each country. Second, I compute a global measure of credit-to-private sector by aggregating country level credit-to-private sector figures again weighted by the level of financial assets of each country. Finally, the global adjusted leverage measure is equal to the negative global leverage innovations when global credit is less than a threshold value and equal to global leverage innovations otherwise. The threshold is set at the 25th percentile of a rolling window on the global credit series. This adjusted leverage measure is directly related to business cycles. Consistent with theoretical and empirical work suggesting that the marginal value of wealth of a financial intermediary is pro-cyclical (see, e.g. Brunnermeier and Pedersen (2009) and Adrian and Shin (2010)), adjusted leverage is positively correlated with changes in real GDP, capital formation, and total factor productivity (TFP) at a country level. In addition, it predicts financial crises and future levels of durables and non-durables consumption. The relation between adjusted leverage and business cycles implies that it is an economically meaningful measure summarizing various aspects of economic activity related to financial intermediaries marginal value of wealth. Based on these properties I argue that this measure can be employed as a reasonable proxy for the SDF of financial intermediaries. Using a single factor model, I perform cross-sectional asset pricing tests across a set of international asset classes. I find that excess returns of currency portfolios and international equity portfolios can be explained by their exposure to global adjusted leverage. More specifically, the global adjusted leverage factor appears with a significantly positive price of risk consistently across all test assets. The global adjusted factor model outperforms benchmark models (see, e.g. Lustig et al. (2011), Menkhoff et al. (2012a), and Menkhoff et al. (2012b)) which aim to explain the cross-section of currency portfolios, and performs similarly to standard multifactor models, such as the Fama-French global three factor plus momentum model, which aim to explain the cross-section of international equity portfolios. The positive price of risk across asset classes is consistent with the theoretical framework developed in this paper where assets that co-vary with intermediaries SDF are associated with a higher risk premium. My findings suggest that the marginal value of wealth of financial intermediaries is indeed an important determinant of asset prices. 3

5 A common criticism of cross-sectional asset pricing tests is that mis-estimated exposures (betas) in the time-series regressions could be explaining a spurious relationship in the cross-section of returns (see, e.g. Lewellen et al. (2010)). I address many of the concerns voiced in Lewellen et al. (2010) by conducting a number of robustness checks. First, I estimate the exposures of test portfolios on the global leverage factor and find that the adjusted leverage betas increase in a pattern consistent with an increasing adjusted leverage being associated with higher premia. Second, I construct an adjusted leverage factor-mimicking portfolio and repeat the asset pricing tests using a longer time-series. The price of risk of the global adjusted leverage factor remains significantly positive across all test assets with the exception of momentum portfolios. Third, I perform beta sorts to address the potential criticism that my results only hold for portfolios used in the tests. I estimate the exposure of country-level market portfolios and country-level financial sector portfolios on the adjusted leverage factor-mimicking portfolio and measure the spread in average returns of these portfolios. The resulting spread is positive, suggesting that the adjusted leverage factor is truly priced in the cross-section. Finally, for all cross-sectional asset pricing tests, I measure the fraction of instances where a randomly generated factor achieves an explanatory power higher than and pricing error lower than that generated by the global adjusted leverage factor. I find that for the majority of test assets this fraction is extremely low (less than 1%). A number of factors aiming to capture the marginal value of wealth of financial intermediaries have been proposed in the literature. Adrian et al. (2014) propose a single-factor intermediary SDF. The factor is a time-series of the shocks to the leverage of securities broker-dealers and carries a large and significantly positive price of risk. On the other hand, He et al. (forthcoming) propose as a two-factor intermediary SDF. The first factor is the market and the second is a time-series of the shocks to the equity capital ratio of primary dealer counterparties of the New York Federal Reserve. Using an extensive set of test assets the authors show that it carries a consistently positive price of risk. I compare the explanatory power of the global adjusted leverage factor against that of the factors proposed by Adrian et al. (2014) and He et al. (forthcoming). I find that the global adjusted leverage factor appears with a consistently positive price of risk across all test assets and that it outperforms both the leverage factor and the capital factor in the cross-section currencies and global equities. The better performance of the adjusted leverage factor as compared to that of other factors proposed in the literature implies that this factor is a more accurate state variable reflecting global financial intermediaries SDF. This paper is organized as follows: Section 2 discusses the related literature and the contribution of the paper; Section 3 develops a theoretical framework that motivates the empirical work; Section 4 presents the data sources; Section 5 presents the construction of the adjusted leverage factor and discusses its properties; 4

6 Section 6 discusses the empirical methodology and how it relates to theory; Section 7 shows the main findings of the paper; Section 8 revisits the empirical findings under alternative theoretical frameworks, compares the performance of global adjusted leverage against other measures proposed in the literature, and decomposes global adjusted leverage into a non-u.s. and a U.S. only component; and finally Section 9 concludes the paper. 2 Related Literature and Contribution This paper is closely related to two main streams of the literature. First, a large stream of literature studies the relation between financial intermediaries and asset prices. 7 Financial institutions are the class of investors whose characteristics most closely align with those of a representative investor in traditional asset pricing models, and thus the study of their marginal value of wealth is expected to provide a more instructive stochastic discount factor (SDF). 8 Models of intermediary-based asset pricing link the marginal value of wealth to intermediaries funding constraints implying that marginal utility is high when funding constraints are binding (see, e.g. Brunnermeier and Pedersen (2009), Geanakoplos (2010), Gromb and Vayanos (2002), and Shleifer and Vishny (1997)). A common theme across these models is a pro-cyclical intermediary leverage, which implies a positive price of risk. 9 In line with this stream of research, Gabaix and Maggiori (2015) develop a model of exchange rate determination based on capital flows in financial markets with frictions. Empirically, Adrian and Shin (2010) show that financial intermediaries adjust their leverage actively according to economic conditions resulting in pro-cyclical leverage. Adrian et al. (2014) and He et al. (forthcoming) show that shocks to the leverage and capital ratios of financial intermediaries, respectively, explain a large portion of the cross-sectional variation in the expected returns of an array of asset classes. 10 Finally, DellaCorte et al. (2016a), focusing on the currency market, show that a global imbalance risk factor (see, e.g. Gabaix and Maggiori (2015)) explains the cross-sectional variation in currency excess returns. The contribution of this paper to the financial intermediation literature is twofold. On the theory side, I deviate from the intermediary-based asset pricing models mentioned above by introducing information 7 Financial intermediaries play a central role in modern markets. The importance of this role and the need for additional research has been part of past AFA presidential addresses (see, e.g. Allen (2001), Duffie (2010), Cochrane (2011)). 8 This approach is in contrast to conventional consumption-based asset pricing models where the marginal investor is the household (see, e.g. Campbell and Cochrane (1999) and Bansal and Yaron (2004)). Households exhibit limited stock market participation (see, e.g. Vissing-Jørgensen (2002)), pay higher transactions costs, and exhibit a lack of financial sophistication (see, e.g. Calvet et al. (2007)). 9 On the other hand, He and Krishnamurthy (2013) and Brunnermeier and Sannikov (2014) propose a central role for intermediaries wealth and generate a countercyclical intermediary leverage (negative price of risk). 10 Etula (2013), Adrian et al. (2015), Adrian et al. (2013) show that the risk-bearing capacity of U.S. securities broker-dealers is a strong predictor of asset returns (commodities, currencies, equities, and bonds). 5

7 production by the agents in the economy as in Gorton and Ordoñez (2014) and Gorton and Ordoñez (2016). More specifically, I propose an alternative mechanism where information production in the economy determines the amount of credit outstanding in the economy and subsequently the level of financial leverage and the SDF of financial intermediaries. On the empirical side, I expand the focus from U.S. to international financial intermediaries, and propose and test the asset pricing properties of an empirical proxy for the SDF of a global financial intermediary. This measure captures multiple aspects of economic activity ranging from capital formation to financial crises and future consumption. The paper establishes an economically meaningful link between the marginal value of wealth of a global financial intermediary and asset prices, thus relating asset prices to the macroeconomy through the financial intermediaries pricing kernel. I deviate from previously used methodologies in both the dimensions of variable construction and scope of test assets. The measure of adjusted leverage is computed by aggregating granular (balance sheet) information in tandem with information on credit-to-private sector from a wide array of countries. This method allows me to obtain a more accurate representation of the SDF of financial institutions on a global level, since I combine two pieces of information (balance sheet information and aggregate credit conditions), which leads to a greater explanatory power over the cross-section of returns. 11 Second, another large stream of literature studies international assets excess returns. The literature around currency excess returns has focused on portfolio strategies based on currency characteristics, such as the interest rate differential (carry trade (see, e.g. Hansen and Hodrick (1980), Meese and Rogoff (1983), Fama (1984), Koijen et al. (2016))), past returns (momentum and value (see, e.g. Menkhoff et al. (2012b) and Asness et al. (2013))), and global foreign exchange volatility. Explanations for currency premia include aggregate consumption growth risk (see, e.g. Lustig and Verdelhan (2007)), currency crash risk and peso problems (see, e.g. Brunnermeier et al. (2008), Burnside et al. (2011a) and Burnside et al. (2011b)), global risk (see, e.g. Lustig et al. (2011)), habits (see, e.g. Verdelhan (2010)), and rare disasters (see, e.g. Farhi and Gabaix (2016)) 12 The literature around international equity excess returns has focused on documenting the existence of size, value, and momentum premia in equity markets across the world (see, e.g. Griffin (2002) and Fama and French (2012)). Global versions of the Fama French three-factor model can explain a large part of variation in international equity excess returns (see, e.g. Fama and French (2012)). Alternative explanatory factors related to funding constraints of investors in international financial markets explain cross-country 11 Prior empirical research uses as proxies for the marginal value of wealth of financial intermediaries the changes in the leverage ratio (see, e.g. Adrian et al. (2014)), or a measure of the intermediary capital ratio (see, e.g. He et al. (forthcoming) without taking into account the level of the available credit in the economy. The importance of high leverage or low available capital varies with respect to the available level of credit in the economy. The marginal utility of an intermediary when both leverage and credit are high is not the same as when leverage is high and credit is low. In the second case the marginal utility of an intermediary is higher. 12 Additional explanations include the term structure (see, e.g. Bansal (1997), country-specific characteristics such as per-capita GDP and inflation (see, e.g. Bansal and Dahlquist (2000)), currency volatility (see, e.g. Menkhoff et al. (2012a) and DellaCorte et al. (2016b)), downside risk CAPM (see, e.g. Lettau et al. (2014)), and global imbalances (see, e.g. DellaCorte et al. (2016a)). 6

8 variation in equity premia (see, e.g Goyenko and Sarkissian (2014) and Malkhozov et al. (2017)). This paper contributes to the international finance literature and more specifically to the above mentioned literature on risk factors associated with risk premia in currency and equity markets. I propose an alternative explanation based on the role of financial intermediaries as marginal investors in these markets. This explanation is based on an economically meaningful link between the marginal value of wealth of a global financial institution and excess returns in the cross-section of currency and international equity portfolios. 3 Theoretical Framework In this section, I develop a two period general equilibrium framework in the spirit of Gorton and Ordoñez (2014) and Gorton and Ordoñez (2016). 3.1 Setting The economy comprises three agents, each with a mass 1 firms, financial institutions, and households and two types of goods capital (numeraire) and land. All agents are risk neutral, apart from the financial institutions which are risk neutral with regard to their lending activities and risk averse with regard to holding firms equity. As in Gorton and Ordoñez (2014) only firms have access to managerial labor (L ), which combined with numeraire (K) produce more numeraire (K ). The production process is stochastic with Leontief technology: A min{k, L } with prob. q K = 0 with prob. (1 q), where A is a parameter determining output when production process is successful and q is the probability that the production process is successful. For the purposes of this model I interpret q as the level of technological innovation. A and q combined describe the efficiency of the production process. Production is efficient (qa > 1) which means that the optimal amount of numeraire is K = L. In this economy, households begin with an endowment of numeraire K > K which can sustain optimal production. Financial institutions are the sole owners of firm equity, which makes them the de-facto marginal investors. Finally, firms own land and are endowed with numeraire in period 1 (K 1 ) but have no means of capital in period 2. Land is not used in production however it derives its value from the amount of numeraire it 7

9 produces at the end of the second period. If land is good, it yields C units of numeraire at the end of the second period; if it is bad, it does not yield anything. Only a fraction ˆp of land is good. In this economy, output and technological innovation (q) are non-verifiable, but the quality of land is not. This makes land valuable as collateral. To receive capital necessary for production, firms pledge a fraction of land as collateral for the loan they receive from the financial institution. This collateral is pledged in turn by financial institutions to facilitate deposits from households. In this setting, C > K which means that land that is good can support the optimal capital size (K ). In period 1 the agents form beliefs about the fraction of land that is of good quality. To determine the true quality of land with certainty, households must pay γ h, while financial institutions must pay γ b, where γ h > γ b. This reflects the fact that financial institutions have superior technology compared to that of households in determining the quality of collateral. 3.2 Optimal Loan for a Single Firm Firms choose between debt that causes information production about the collateral leading to informationsensitive debt, and debt that does not induce information production leading to information-insensitive debt. Information acquisition for the financial institution bears a cost γ b. As in Gorton and Ordoñez (2014), I determine conditions under which debt is information-sensitive or information-insensitive Information-Sensitive Debt In this case, financial institutions discover the true value of the firms land at a cost γ b. Financial institutions are risk neutral when it comes to their lending activity which means that they break even: p(qr b IS + (1 q)x b ISC f K b ) = γ b, (1) where K b is the actual loan from the financial institution to the firm, R b IS is the face value of the debt, and x b IS is the fraction of land posted as collateral. The fraction of collateral that a firm posts is determined by, R b IS = x b ISC f x b IS = pkb + γ b pc f If R b IS > xb IS Cf, the firm would always hand over the collateral instead of repaying the loan. On the other hand, if 8

10 Expected profits (net of land value) are E(π p, IS) = p(qak b x b ISC f ) = pk (qa 1) γ b. 14 (2) Information-Insensitive Debt In this case, financial institutions do not produce information regarding the quality of firms land. As financial institutions are risk neutral with respect to their lending activity and break even, qr b II + (1 q)px b IIC f = K b, (3) where RII b = pxb II Cf as with the previous case. Financial institutions could potentially deviate and privately check the quality of the land prior to lending capital. The following condition guarantees that they will not deviate since the expected payoff from producing information is less than the cost (γ b ): p(qr b II + (1 q)x b IIC f K b ) < γ b (1 p)(1 q)k b < γ b. The financial institution lends the optimal amount of capital (K ) if the above condition is satisfied for K b = K. However, if the above condition is not satisfied, the amount of capital is K b = if collateral value is low. Combining the above, the loan level for information-insensitive debt is: γ b (1 p)(1 q) or pcf, { K b (p, q II) = min K γ b }, (1 p)(1 q), pcf. (4) Expected profits (net of land value) are E(π p, II) = pqak b x b IIpC f = K(p, q II)(qA 1). (5) Equating profits under information-sensitive debt (equation 2) with profits under information-insensitive debt (equation 5) allows to pin down the level of the loan under information-sensitive debt: R b IS < xb IS Cf the firm would always sell the collateral and repay the loan. This means that R b IS = xb IS Cf. 14 I assume that it is feasible to borrow the optimal amount of capital (K ), which means that x b IS = pkb +γ b pk (qa 1) > γ b. Combining the two yields the condition qa < C f /K. pc f 1 and that 9

11 K b (p, q IS) = pk γb qa 1. (6) 3.3 Optimal Deposits for a Financial Institution In this setting the owners of capital are households, which deposit their wealth in financial institutions, which in their turn lend it to firms. The banks choose between deposits that cause information production about the ability of the bank to repay, and deposits that do not induce information production. Both banks and households make their decisions simultaneously which means that the household cannot infer the quality of collateral from observing the bank s loan. The creditworthiness of the financial institution is determined by the amount of collateral that the firm has pledged for the loan it received. The cost of information acquisition for the household is γ h Information-Sensitive Deposits In this case, households discover the true value of the financial institution s loans, backed by firms land as collateral, by incurring the cost γ h. Households are risk neutral and break even: p(qr h IS + (1 q)x h ISC b K h ) = γ h (7) where K h is the actual amount of deposits in the financial institution, R h IS is the face value of deposits, and x h IS is the fraction of the financial institution s loans posted as collateral. For the same reason as before, the fraction of collateral that the financial institution posts is x b IS = pkh +γ h. Since the collateral posted by the financial institution is the land that has been posted by the firm to obtain its loan, C b = C f. pc b Expected profits for the financial institution are E(π b p, IS) = px b ISC f px h ISC b = pk b + γ b pk h γ h. 15 (8) Information-Insensitive Deposits In this contract, financial institutions attract deposits without triggering production of information regarding their assets. Since households are risk neutral and break even: 15 Because of the assumption that γ h > γ b the bank will always produce information before the household does as both q and p decline. 10

12 qr h II + (1 q)px h IIC b = K h, (9) where R h II = pxh II Cb for the same reasons as before, which means that x h II = pkh +γ h pc b. For the contract to be information-insensitive, no household should have an incentive to deviate. This is guaranteed by: p(qr h II + (1 q)x h IIC b K h ) < γ h (1 p)(1 q)k h < γ h. As with the information-sensitive loan to the firm, the deposits contract will reach the optimal amount of capital (K ) if the above condition is satisfied. K h = If it is not, the amount of deposits will either be γ h (1 p)(1 q), if the financial institution faces credit constraints, or pcb if the collateral value is low. Thus, { K h (p, q II) = min K γ h }, (1 p)(1 q), pcb. (10) Expected profits for the financial institution are E(π b p, II) = px b IIC f px h IIC b = K b K h. (11) Equating profits under information-sensitive deposits (equation 8) with profits under informationinsensitive deposits (equation 11) allows to pin down the level of deposits under information-sensitive debt: K h (p, q IS) = (1 p)kb IS + γh γ b. (12) 1 p Figure 1a shows the amount of credit in the form of deposits and loans that the household and the financial institution respectively are willing to make depending on the probability of success of the production process (q) keeping the fraction of land that is good (p) constant. For the remainder of this section, this probability will represent technological innovation. The cutoffs in Figure 1a are determined as follows: Cutoff A occurs at the level of technological innovation below which firms reduce their borrowing so that they do not induce information production. From above: 11

13 K γ b = (1 p)(1 q) γ b qb,h II = 1 K (1 p). (13) Cutoff B is determined by the level of technological innovation below which financial institutions reduce their deposits thus avoiding information production from households. As before: K = γ h (1 p)(1 q) γ h qh,h II = 1 K (1 p). (14) Cutoff C is obtained after equalizing information-sensitive debt with information-insensitive debt: γ b (1 p)(1 q) = γ b pk (qa 1). (15) The positive root of the above quadratic equation (qis b ) is the level of technological innovation below which financial institutions acquire information about the quality of land that firms post as collateral for their loan. Cutoff D is obtained similarly for the case of deposits, γ h (1 p)(1 q) = γ b pk (qa 1) + γb γ h (p 1). (16) The above cutoffs create four distinct regions: (1) Between A and B (B(II), H(II)), both the bank and the household are information-insensitive (II). Also, bank credit is constrained and firms cannot borrow without triggering information production, which means that credit rationing takes place in bank lending; (2) Between B and C (B(II), H(II)), both agents are information-insensitive (II) but credit constraints lead to rationing in both bank lending and deposits; (3) Between C and D (B(IS), H(II)), the bank is information-sensitive (IS) and at a cost γ b discovers the true value of the land. The household is still information-insensitive and rationing takes place in deposits; and (4) Below D (B(IS), H(II)), both agents are information-insensitive with banks producing information about the quality of land and households producing information about the quality of the assets of the financial institution. 16 The amount of credit that is available in the economy to fund firms projects is a function of technological innovation (q), the quality of land (p), and the cost of information production for financial institutions (γ b ) and households (γ h ). For the purposes of this paper, I assume that the quality of land and the cost of 16 Additional regions become relevant depending on the level of the fraction of land that is good p and the value C of land. Low collateral value can constrain the amount of deposits and subsequent loan amount, however for high enough levels of p, it becomes irrelevant. 12

14 information acquisition remain constant throughout the two periods. Proposition 1 summarizes the relation between technological innovation and the amount of credit in the economy. The proof is trivial. Proposition 1. (Effect of technological innovation on credit.) For fixed values of γ b, γ h, and p, with γ b < γ h and K < pc f = pc b : Deposits are an increasing function of technological innovation (q) for q < q h,h II otherwise. Loans are an increasing function of technological innovation (q) for q < q b,h II otherwise. and independent of q and independent of q 3.4 Firm Valuation As mentioned above, in this economy the financial institutions are the sole owners of firms. They are risk averse with respect to their equity holdings and risk neutral with respect to their lending activity. This assumption is motivated primarily by current banking regulation which mandates bank capital requirements and determines internal risk-management policies. More specifically, according to the Third Basel Accord (Basel III) equities are deemed substantially less liquid than high quality corporate debt. This means that equities in intermediaries balance sheets require a higher capital provision compared to that required for high quality corporate debt. 17 I derive the stochastic discount factor (SDF) and compute the financial leverage of financial institutions in a two period general equilibrium framework. Table 1 summarizes the timeline for this economy. The financial institution maximizes a logarithmic utility function with two terms, a deterministic component for the first period and a stochastic component for the second period, max Eu(C 1, C 2 ) = log(c 1 ) + βelog(c 2 ) (17) {C t} 2 t=1 Subject to: C 1 + K b + γ b + V 1 a 1 = (π 1 (q 1 ) + V 1 )a 0 + K h (18) and K h + C 2 = K b + π 2 (q)a 1 + pc f (19) 17 Basel III requires financial institutions and non-bank financial companies deemed systemically important to have enough high-quality liquid assets (HQLA) which can be quickly liquidated to meet possible future liquidity needs. Assets are classified into three groups (Level 1, Level 2A, and Level 2B) according to their liquidity properties. A total HQLA is computed as the weighted sum of the the asset value times a weight that is consistent with its liquidity group. Haircuts vary from 0% for assets in Level 1 to 50% for assets in Level 2B. Common equity falls into Level 2B and is subject to a 50% haircut which is significantly higher compared to the 15% haircut of high quality (>AA- rating) corporate debt (see, e.g. and 13

15 where C t is the period t consumption, K b the amount of the loan to the firm, K h the amount of deposits in the financial institution, γ b the cost of information acquisition for the financial institution, a t the time t fraction of firm that is held by the financial institution, p the fraction of land that is good, C the numeraire that land that is good delivers at the end of period 2, and π t (q) the period t firm profits as a function of technological innovation (q). 18 Market clearing requires that a 0 = 1, a 1 = 1, C 1 +K b +γ b = π 1 (q 1 )+K h, and K h +C 2 = K b +π 2 (q)+pc f. K h and K b are determined above. First order conditions with respect to a 1 yield: ( V 1 = E β C ) 1 π 2 (q) C 2 (20) which means that the SDF for the financial institution is: m = Kh K b + π 1 (q 1 ) γ b K b K h + π 2 (q) + pc f (21) Financial leverage is defined as the ratio of assets to assets minus liabilities: l = K b + V 1 K b + V 1 K h (22) The relation between a financial institution s SDF and its financial leverage depends on the level of information acquisition from the household and the financial institution, which in turn is directly related to the level of technological innovation that the economy experiences. Technological innovation is also directly related to the level of credit outstanding in the economy (Proposition 1). Hence any relation between a financial institution s SDF and its financial leverage across levels of technological innovation holds across levels of credit outstanding. Proposition 2 summarizes this relation for different regimes of information production defined by the level of technological innovation. 19 Proposition 2. (Stochastic discount factor and financial leverage.) The relation between the SDF of a financial intermediary and its financial leverage depends on the level of technological innovation (q) and information acquisition from the household and the financial intermediary: 18 π 1 (q 1 ) = Aq 1 K 1, and π 2 (q) = K b (p, q II)(qA 1) when bank loans are information-insensitive and π 2 (q) = pk (qa 1) γ b when bank loans are information-sensitive. 19 The proof can be found in the appendix. 14

16 Bank loans and household deposits are information-insensitive with no credit constraints present (B(II), H(II)): Leverage is constant and SDF a negative function of technological innovation. Bank loans and household deposits are information-insensitive, but credit constraints are present (B(II), H(II)): Leverage is positively and SDF negatively correlated with technological innovation. Bank loans and household deposits are information-insensitive, but deposit and credit constraints are present (B(II), H(II)): Leverage is positively and SDF negatively correlated with technological innovation. Bank loans are information-sensitive and household deposits are information-insensitive with deposit constraints present (B(IS), H(II)): Both leverage and SDF a positive function of technological innovation Bank loans and household deposits are information-sensitive (B(IS), H(IS)): Leverage is positively and SDF negatively correlated with technological innovation. Figure 1b provides an illustration of Proposition 2. For a high level of technological innovation and credit outstanding, the relationship between SDF and financial leverage is negative. An increase in financial leverage is associated with a decrease in the SDF of the financial institution. This usually represents times of economic growth where the financial institution is able to sustain a relatively high level of leverage due to the availability of a large number of profitable investment opportunities and the high level of funding it secures from the household. However, when technological innovation and credit outstanding is at a relatively low level and the household does not yet produce information, the relationship between the two variables is the opposite. 20 An increase in financial leverage is associated with an increase in the SDF of the financial institution. The change in the relationship between the two variables is caused by the fact that the rate of increase in deposits is higher than that in loans and the valuation of the firm. The increase in funding from households does not keep up with the improvement in the profitability of investment opportunities in the economy leading to an increasing leverage ratio and an increasing SDF. This usually represents times of economic recession or financial crises where the financial institution may have sufficient funding, but not enough investment opportunities leading to high levels of leverage. Both Proposition 2 and Figure 1b underline the point that financial leverage is not always an accurate measure of the SDF of a financial institution. In times when credit is low the relation between the two variables reverses and a more accurate proxy of the marginal value of wealth of financial institutions should take that into account. When credit is high, assets that co-vary with leverage are riskier and as a result should earn higher premia. On the other hand, when credit is low, assets that co-vary with leverage are less 20 Usually the household does not produce information about the quality of the assets of the financial institution, unless there is widespread uncertainty about the health of the financial system. Which makes the last case of Proposition 2 less likely to be observed in the data. 15

17 risky and should earn lower premia. 21 In the following section, I propose an empirical measure that takes this point directly into account and combines information from both credit and financial leverage. 4 Data This section describes the data used in the empirical analysis. I provide details on the construction of the adjusted leverage factor, currency excess returns, currency portfolios, and the macroeconomic variables used in the empirical tests. 4.1 Leverage Ratio Intermediary asset pricing models suggest that financial intermediaries are sophisticated investors who play a leading role in capital markets. They are considered marginal investors which means that their pricing kernel is relevant for pricing the cross-section of risky assets. 22 Motivated by the theoretical framework developed above, I construct a proxy for the marginal value of wealth which takes into account the leverage ratios of financial intermediaries and the credit conditions in the economy. I compute leverage ratios for financial institutions that act as broker-dealers for an international panel (Table 2) using data from Thomson Reuters WorldScope. 23 I delete duplicate entries and data on ADRs. Leverage ratios are computed as follows: ( ) Assets Leverage = log Assets Liabilities (23) 4.2 Assets Portfolios For the purposes of the asset pricing tests of Section 7.1 I use currency and global equity portfolios. I construct six forward discount portfolios following the methodology of Lustig et al. (2011). I rank 21 This relation can reverse if financial institutions face a higher cost of information acquisition on the firms assets than households do (γ b > γ h ). However, this does not seem to be the case in modern economies where financial institutions have a plethora of resources at their disposal to research and identify the quality of the posted collateral. 22 There are two approaches to modeling an intermediary pricing kernel each of which has a different theoretical motivation. The first uses intermediary leverage as a proxy for the SDF (see, e.g. Brunnermeier and Pedersen (2009), while the second uses intermediary wealth (see, e.g. He and Krishnamurthy (2013) and Brunnermeier and Sannikov (2014)). Throughout this paper I follow the first approach. 23 The actual WorldScope industry codes for financial firms used in the analysis are: 4310 for Commercial Banks - Multi-Bank Holding Companies (used only for international financial institutions), 4394 for Securities Brokerage Firms, and 4395 for Miscellaneous Financial. Financial institutions classified as commercial banks in the database serve as broker-dealers in many countries. Commercial Banks - Multi-Bank Holding Companies (4310, U.S only data), Commercial Banks - One Bank Holding Companies (4320), Investment Companies (4350), Commercial Finance Companies (4360), Insurance Companies (4370), Land and Real Estate (4380), Personal Loan Company (4390), Real Estate Investment Trust Companies, including Business Trusts (4391), Rental & Leasing (4392), and Savings & Loan Holding Companies (4393) are not included. 16

18 currencies from low to high interest rates such that portfolio 1 contains currencies with the lowest forward discounts, and portfolio 6 contains the currencies with the highest forward discounts. The strategy that is long on portfolio 6 and short on portfolio 1 represents carry trade and constitutes the carry factor (CAR) that I use in the following asset pricing tests. Currency momentum portfolios are constructed using the methodology of Menkhoff et al. (2012b). Each month I form six portfolios on the basis of excess currency returns of the previous n months. Portfolio 1 contains currencies with the lowest prior n month returns, while portfolio 6 comprises of currencies with the highest prior n month returns. I construct two types of momentum portfolios, long-term momentum (n = 12 months) and short-term momentum (n = 1 month). The strategy that is long on short-term momentum portfolio 6 and short on portfolio 1 constitutes the momentum factor (MOM). I construct value portfolios following the methodology in Asness et al. (2013). As with currency momentum, I form six portfolios based on the lagged five-year excess return of the currency of each country in the sample. I assign the lowest lagged returns to portfolio 1 and the highest to portfolio 6. Global equities portfolios comprise twenty-five international size and value sorted portfolios and the twenty-five international size and momentum portfolios all obtained from Kenneth French s online data library Macroeconomic Data In Section 5, I explore the properties of the adjusted leverage measure with respect to macroeconomic variables. Annual Real GDP and capital formation are from the Penn World Tables (PWT), domestic credit to private sector, credit to households and credit to corporates are from the World Bank World Development Indicators. Financial crisis episodes are from Valencia and Laeven (2012). Global imbalances are defined as the difference between assets and liabilities denominated in the same currency. I construct the measure of total global imbalances, which is the sum of global imbalances issued in domestic and foreign currency standardized for the GDP of a country. For additional details see Bénétrix et al. (2015). 25 Market returns are the average market returns at a country level, financial intermediaries returns are the average financial sector returns at a country level, and global volatility is the average market volatility among an international set of countries (see, e.g. Chousakos et al. (2016)). In addition, and only for the U.S. economy, I collect data on credit spreads, per capita consumption on durables, non-durables, and private investment from the Federal Reserve Economic Data (FRED) maintained by the St. Louis FED. Table 3 summarizes the data used in this paper The dataset can be found at Philip Lane s website 17

19 5 Adjusted Leverage In the literature the marginal value of wealth of financial institutions is proxied by their financial leverage and more specifically by seasonally adjusted changes in the level of broker-dealer leverage (see, e.g. Adrian et al. (2014)). This could be potentially problematic because as discussed in Section 3 financial leverage or its reciprocal capital ratio does not fully characterize the risk-bearing capacity of financial intermediaries across all states of the economy. An increase in financial leverage does not necessarily indicate a decrease in the marginal value of wealth for the financial institution. This could be true in times of economic growth, where credit outstanding is high and the financial system can sustain high levels of leverage as financial intermediaries are able to pursue profitable investments and at the same time raise capital if needed. However, this is not the case in times of recessions or financial crises, where the amount of credit outstanding is low. According to the theoretical framework discussed above, it possible that in a low credit environment an increase in leverage is associated with an increase in the marginal value of wealth of the financial institution due to a lack of profitable investment opportunities. A number of empirical findings, summarized below, suggest that financial leverage interacts differently with key characteristics of financial intermediaries depending on the level of credit outstanding in the economy. The level of assets of financial intermediaries reflects among other factors their risk-bearing capacity. A high level of assets in intermediaries balance sheets is usually the result of higher risk-bearing capacity. If the financial leverage of an intermediary is a good proxy for its risk-bearing capacity, then the relation between leverage and the level of future assets is expected to be positive across all states of the economy (see e.g. Adrian et al. (2014)). The following regression specification provides a test of the above claim: assets n,t+q = α n + α t+q + β X n,t + γassets n,t + ɛ n,t+q (24) where assets n,t+q is the logarithm of the financial assets of country n at time t + q quarters, X n,t = (fin.leverage n,t, fin.leverage n,t 1(Credit n,t > 75%), fin.leverage 1(Credit t < 25%)), fin.leverage n,t is the logarithm of financial leverage of country n at time t, 1(Credit n,t > 75%) is a dummy variable representing instances where credit-to-private sector of a given country is higher than the 75th percentile of the cross-section of credit outstanding values at a given time, 1(Credit n,t < 25%) is a dummy variable representing instances where credit-to-private sector of a given country is lower than the 25th percentile of the cross-section of credit outstanding values at a given time, α n represents country fixed effects, and α t+q time fixed effects. 18