Real Interest Rates, Transport Costs and Trade

Transcription

1 Real Interest Rates, Transport Costs and Trade Thomas Baranga IR/PS, UCSD Sept 3rd, 2013 Abstract International trade takes more time, and so requires more working capital, than domestic transactions. This may make trade particularly sensitive to financial conditions. The effect of finance on international trade can be identified by examining the tilting of trade, as firms facing higher finance costs economise on working capital by selling relatively more to nearer than further markets. Evidence of this phenomenon is found in a cross-country panel of aggregate trade flows, and by comparing the import elasticities of US commodities that are transported by boat with those flown by plane. 1 Introduction The Great Recession of ushered in an extraordinary collapse in world trade, at least four aspects of which are remarkable: the trade collapse was global (the 50 largest exporting and importing countries all saw declines in their total merchandise trade in 2009, mostly by double digits, with a similar pattern for trade in services 1 ; trade fell across all goods categories 2 ; trade fell by much more than GDP (world GDP fell by 2.5% in 2009 while trade fell by 12% 3, and by 33% Q Q ); and world trade flows Baranga: IR/PS 0519, University of California - San Diego, 9500 Gilman Drive, La Jolla, CA ( tbaranga@ucsd.edu; ). 1 WTO (2010, Table 1.8, p.13; Table 1.10, p.15) 2 WTO (2010, p.36, p.39) 3 WTO (2010, Table 1.1, p.8) 4 WTO (2009, p.3) 1

2 subsequently recovered strongly from the crisis (world merchandise exports increased by 14% in 2010, while world GDP expanded by 3.5% 5 ). Several theories have been advanced to explain the largest shock to global trade since the Great Depression: the increased fragmentation of production across national borders and greater trade in intermediate goods has increased the elasticity of trade s response to GDP shocks 6 ; the composition of the negative demand shock, falling particularly on durable consumer goods such as automobiles, had particularly severe consequences for trade, in which consumer durables are overrepresented relative to GDP 7 ; and the obvious origins of the economic crisis in financial markets has led to renewed focus on the role of finance in facilitating international trade, and the effects a tightening of credit conditions could have on trade flows 8. There has been a vigorous debate over the sources of trade s excess volatility, from which a clear consensus has not yet emerged (Baldwin (2009) provides excellent summaries of the leading points of view). Several authors have weighed the competing hypotheses and rejected a role for financial factors in accelerating the trade collapse (beyond triggering the initial collapse in production and demand) 9 Among studies finding an effect, there is also disagreement over whether it is felt on the intensive or extensive margins of trade: Behrens et al. (2010) and Bricongne, Fontagné, Gaulier, Taglioni & Vicard (2012) find that only the intensive margin of trade was affected during the crisis for Belgian and French firms respectively, while Berman & Héricourt (2008) and Muûls 5 WTO (2011, p.14) 6 Freund (2009) estimates this elasticity had risen to 3.69 by the 2000s, and is larger in downturns. Bems, Johnson & Yi (2010) estimate an elasticity of 2.8, which could account for 70% of the fall in trade. 7 Eaton, Kortum, Neiman & Romalis (2011) estimate that 80% of the fall in the trade/gdp ratio can be explained by a shift in demand away from manufactures, while Alessandria, Kaboski & Midrigan (2010) argue that the destocking of intermediate inputs from inventories can quantitatively explain the overreaction of trade relative to GDP. 8 Leading contributions both theoretical and empirical include Iacovone & Zavacka (2009), Auboin (2009), Amiti & Weinstein (2011), Ahn, Amiti & Weinstein (2011), Ahn (2011), Chor & Manova (2011), Feenstra, Li & Yu (2011), Paravisini, Rappoport, Schnabl & Wolfenzon (2012), Berman, De Sousa, Martin & Mayer (2012), Schmidt-Eisenlohr (2012) and Antràs & Foley (2013). 9 In addition to Eaton et al. (2011) and Alessandria et al. (2010), which claim to explain 70%-80% of the trade collapse through non-financial channels, Levchenko, Lewis & Tesar (2010a, p.248) find some support for compositional effects and vertical linkages but do not detect any impact of trade credit on the reduction in international trade, and Levchenko, Lewis & Tesar (2010b, p.3) conclude that evidence of financial factors has proven hard to find. Behrens, Corcos & Mion (2010) find that the state of Belgian firms balance sheets had only a modest impact on their international transactions, and did not disrupt international trade disproportionately to total turnover. 2

3 (2012) find effects only on the extensive margin for firms in emerging markets and Belgian firms. Paravisini et al. (2012) and Berman et al. (2012), in the setting closest to this paper, find a significant impact on both margins. A wide range of empirical strategies have been employed in the literature. Some studies compare trade flows before and during established crises periods, where additional sources of variation could be cross-country case studies (Ronci (2004), Freund (2009)), or time-to-market Berman et al. (2012)); or differential dependence by sector on external finance (Iacovone & Zavacka (2009), Levchenko et al. (2010a), Levchenko et al. (2010b), Chor & Manova (2011)) or mode of transport (Ahn et al. (2011)); or by firm (where firm-level exposure to financial shocks may be inferred from balance sheet information, Behrens et al. (2010), or the health of linked banks, Paravisini et al. (2012)). Other studies have explored the relationship in less turbulent settings over a longer time period, again exploiting different sources of variation in exposure to finance: cross-country net interest margins and time-to-market in the case of Schmidt-Eisenlohr (2012), or firms relationships with banks (Amiti & Weinstein (2011)) or interest payments (Feenstra et al. (2011)). This paper also takes a long view, with an aim to estimating the elasticity of trade to a price measure of the cost of finance, real interest rates, by using cross-country variation in the time-to-market and sectoral variation in the mode of transport to identify effects on both intensive and extensive margins. The cost of capital emerges as an important component of trade costs even under normal business conditions, with a significant impact both on whether there is any trade between two markets, and the value of exports and imports if positive. The main empirical difference between this paper and others using crosscountry variation is the choice of a measure of real, rather than nominal, cost of capital. In a large cross-country panel that includes many emerging markets, there are frequent instances of nominal interest rates in excess of 40%; but correction for the off-setting levels of high inflation suggest that the true cost of finance facing local firms was actually more modest. While a cross-country specification using aggregate trade flows may seem too blunt an approach to tease out the mechanism by which financial costs are having their effect, it has important benefits: there is a sufficient density of bilateral trade flows that one can saturate the cross-section with countryspecific dummies to control for all aggregate macroeconomic shocks, while still identifying the effect of variables with bilateral country-pair variation. A simple theoretical model suggests that the time-to-market amplifies the impact of financial frictions, which yields the cross-country variation to test for the role of finance while rigorously controlling for omitted variables bias. Cross-country studies also naturally lend themselves to symmetric treat- 3

4 ment of financial factors in both the importing and exporting countries 10, which is not always true of analysis of firms transactions-level data, as typically the ressearcher does not have symmetric information about the firm s trade partner, a potential source of omitted variable bias 11. The identification yielded by cross-country variation in time-to-market is testable even without transactions level data, by comparing the effects for commodities that are transported by boat to those transported by plane. Finding that the effect is relevant only for slow goods provides strong support that the hypothesised mechanism is indeed in action. 2 Interest Rates and Trade The hypothesis that finance is even more important for international trade than for domestic production stems from the fact that international transactions take longer to complete, and so represent a more intensive commitment of firms working capital than purely domestic sales 12. The variation in time to deliver goods between countries provides the variation to identify a role for the cost of capital, where variation comes both in the physical distance, as well as the speed of transport used for particular commodities (boat versus plane). This variation across bilateral trading routes and commodities can be exploited while controlling for all aggregate macroeconomic shocks to supply and demand in source and destination countries, which would otherwise be a source of endogeneity as they are determined simultaneously along with firms cost of capital. The additional time lag in delivery of international orders implies that working capital is tied up for longer while the goods are in transit. In principle, this could be financed by either the importing or exporting firm, de- 10 As also noted by Schmidt-Eisenlohr (2012). 11 Consider a setting in which one can control for a firm s sector and its balance sheet, but not the financial situation of its trade partner, and one finds that firms exporting intermediate goods experienced a differenially large fall in sales in a financial crisis, independent of their own financial health. This could be interpreted as rejecting a role for financial frictions in accelerating trade collapses in favour of an effect driven by destocking of inventories; but it could also be that the financial vulnerability of firms importing intermediates is systematically different from those importing final goods, and that in the absence of data on the importer s side of the transaction the financial factor loads on the sectoral variables. 12 The role of time-to-market as a significant trade barrier in itself is discussed by Djankov, Freund & Pham (2010) and Hummels & Schaur (2012), and the connection to finance is tested, with mixed results, in Levchenko et al. (2010b), Paravisini et al. (2012), Amiti & Weinstein (2011), Ahn et al. (2011), Schmidt-Eisenlohr (2012) and Berman et al. (2012). 4

5 pending on whether the goods are paid for in advance, or on delivery. Both practices are widespread: in a March 2009 survey of 44 leading banks providing trade finance (FImetrix 2009), banks reported that in Oct 2007, 33% of international trade had bank-intermediated financing (principally in the form of letters of credit), while 48% was financed on open account by exporters being paid after delivery, and 19% by importers paying cash-in-advance. There was little change in this composition over the course of the financial crisis, with the share of bank-intermediated transactions actually rising slightly, to 35% in Oct 2008 and 36% in Jan 2009, while the volume of business fell precipitously (the total value of banks letters of credit fell 8% Oct Oct 2008 and by another 11% Oct Jan 2009; the amount of trade credit fell in every region; and 71% of banks reported a fall in their own issuance of trade credit. 57% of banks reported that the reason for the decline was Less credit available at your own institution. The spread over cost charged by banks for a standby letter of credit rose from 18 to 32 basis points after Oct 2008, and 71% of banks reported that the pricing of trade finance increased due to Own institution s increased cost of funds ) 13. Antràs & Foley (2013) analyse transactions level data for a large US food exporter, and document the use of a wide variety of contractual arrangements (albeit in this particular case significantly fewer are directly intermediated: 42.4% of the value of transactions were on cash-in-advance terms, 41.3% on open account, and only 5.5% by letter of credit, and 10.7% by documentary collection). Antràs & Foley (2013) and Schmidt-Eisenlohr (2012) have both proposed models of optimal trade financing, in which the exporter makes a take-it-orleave-it offer to the importer, based on the relative cost of capital in either country, and the probability of local jurisdictions enforcing contracts, and finds some supporting evidence that the terms of financing respond to these factors 14. While survey evidence suggests that on average the exporter is more likely to bear the financing burden and that there do not appear to be big changes in contracts in response to shocks, the potentially endogenous nature of the financing of trade poses a challenge to empirical specifcations. One approach could be to include symmetric measures of capital costs and let the data tell us whether on average the importer or exporter s financial conditions are more relevant. Another is to explore whether the formal contract matters for the ultimate impact on trade flows. In fact, in a sim- 13 FImetrix (2009) 14 Open Account financing is more likely when exporters can borrow relatively cheaply and the importing firm s jurisdiction has stronger contract enforcement. 5

6 ple model, the incidence of financial frictions is identical, irrespective of who formally finances the transaction. Real interest rates are used as the measure of local firms cost of capital. To my knowledge, real interest rates have not been used in other related studies 15. The motivation for the use of real interest rates is two-fold. The empirical strategy is to estimate the role of financial frictions using a cross-country gravity equation, where trade costs are implicitly modelled ad valorem, which naturally leads to using a price rather than quantity measure of the cost of capital. Secondly, in a large cross-country setting, nominal interest rates may be a very misleading measure of the true cost of time or capital. For example, an Argentine exporter that ships to China and is not paid until after the goods are delivered at the margin may be borrowing at an annualised interest rate of 50% in peso; but if Argentine annualised inflation is running at 75%, and hence on average the peso is depreciating against the RMB at a similar rate, then the exporter actually benefits from a delayed payment if they invoiced in RMB (and if they invoiced in peso, then on average, the agreed peso price will reflect the anticipated depreciated exchange rate at the time of settlement, so that the exporter can still benefit from negative local real interest rates). While the link from inflation to exchange rates is loose in the short-run, over long horizons PPP holds relatively well 16, and on average real interest rates may reflect more accurately local firms effective capital costs when selling overseas as well as locally. The choice of real interest rates is also partly motivated by the recent experience, partly by a glance at the data. The aggressive cutting of nominal short-term interest rates by central banks around the world during the financial crisis is prima facie a problem for a theory trying to relate the price of capital to the trade collapse 17. However, the crisis also led to a sharp, but short-lived, period of deflation in much of the OECD, with the result that real interest rates rose significantly in the period in which trade was in free fall. This relationship between interest rates and trade is long-standing, as Figures (1)-(4) illustrate for both the US and Japan: while there is a clear 15 Ronci (2004) and Berman et al. (2012) map crisis episodes into dummy variables; Amiti & Weinstein (2011) and Paravisini et al. (2012) both use measures of the quantity of credit extended by banks. Studies of the effect of the price of credit, such as Chor & Manova (2011), Levchenko et al. (2010a) and Levchenko et al. (2010b) have used the nominal interbank rate, or in the case of Schmidt-Eisenlohr (2012), the nominal net interest margin. Other studies, such as Behrens et al. (2010) or Feenstra et al. (2011) use accounting measures from firms balance sheets. 16 See Taylor & Taylor (2004) for a survey. 17 Although of course quantity measures might tell a different story, if credit-rationing was widespread. 6

7 positive long-run correlation between nominal interest rates and trade, the correlation between trade and real interest rates is clearly negative. 2.1 Two Models of Trade Finance Risk-Free Trade The cost of financing the transit of international trade is naturally modelled as a component of the iceberg transport costs estimated in a standard gravity equation, such as Anderson & van Wincoop (2003). We assume firms are monopolistically competitive, and face a common elasticity of substitution, σ, for their goods in all markets. We also assume that firms face a choice of risk-free contracts in which either the exporter or importer agrees to bear the burden of financing the transaction until final settlement, and that firms are constrained to financing themselves in their local capital markets. Firms well-known pricing decision is that firm i selling into market j charges a constant mark-up over its marginal cost of delivery to j, yielding demand of p ij = σ σ 1 c ij x ij = p σ ij Pj 1 σ Y j where Y j is income and P j the ideal price index in j. Denote the marginal cost of producing the good for the home market as c i, and the additional costs of delivery to j as τ ij, which might consist of financial (τ f ) and non-financial factors (τ g, e.g. transport costs, tariffs, etc). The non-financial factors are traditionally modelled as ad valorem or iceberg costs, so they raise the cost of supplying j to τijc g i. The financing costs may depend on the terms of the contract. If the exporter finances the deal on open account terms, then at the time the order is placed he incurs working capital costs of c i, amplified to τijc g i to deliver the goods to market, which must be financed (abstracting from additional financial frictions which might drive a wedge between the firm s internal cost of capital and the local market rate) at the local cost of capital, a real interest rate r i. By the time of settlement of the trade, the total cost to the exporter, inclusive of compounded interest, will be (1+r i ) t ij τijc g i, which is of the general form τ ij c i, where the financing cost τ f = (1 + r i ) t ij factors naturally into the general form of the iceberg trade cost, τ ij = τijτ f ij. g 7

8 An exporter financing the transaction will thus charge a price p E ij = σ σ 1 (1 + r i) t ij τijc g i = τ f,e ij τ g ijc i If delivery is financed by an importer in country j, who pays cash-inadvance at the time of the order, then the price paid up front will reflect a mark-up over the exporter s costs of delivery, p I ij = σ σ 1 τ g ijc i However, the effective cost to the importer at the time of settlement is higher, as he has had to fund his cash-in-advance payment out of his working capital, leading to cumulated interest that inflates the true cost to p I ij = σ σ 1 (1 + r j) t ij τijc g i = τ f,i ij τ g ijc i The only difference the terms of the contract make for final demand x ij are which party s interest rate costs are passed on through the price. Aggregation of such a demand system following Anderson & van Wincoop (2003) yields an aggregate gravity equation for bilateral trade of the form X ijt = Y ity jt Y W t ( τijt Π it P jt ) 1 σ where τ ijt = τ f,? ijt τ g ijt, with the financial costs depending on the terms of the contract. This can be estimated in the standard way by log-linearising, and including time-varying importer and exporter dummies for every period to absorb the potentially unobserved macroeconomic supply and demand shocks, Y it and Y jt, as well as the general equilibrium effects expressed through the multilateral resistance variables Π jt and P jt 18. Log-linearisation of the trade costs, assuming a standard model of the non-financial costs such as τ g ij = d ρ ij, where d ij might be a vector including bilateral distance, sharing a common language, etc, yields ln(τ ij ) = t ij ln(1 + r? ) + ρ ln(d ij ) (1) The bilateral variation injected by the interaction between t ij and r implies that the tilting effect of capital costs on trade can still be estimated, 18 Following Feenstra (2004) 8

9 even when dummying out all country-specific macroeconomic variables, including the level of interest rates (macroeconomic variables which might well be endogenous if directly controlled for, or a source of omitted variable bias if omitted). Exporters in a country with high real interest rates should tend to sell more to nearby destinations relative to their exports to more distance markets, while exporters facing low interest rates see trade volumes decay more gradually with distance. This variable tilting of trade would efficiently conserve these firms working capital, and should be perceived if these transit costs are significant enough to be priced into exports. When bringing this to aggregate data, without any information on the actual contracts used to finance trade, one could take an agnostic approach, running a horse race and including both the importer s or exporter s cost of capital to see which the data finds on average to be more significant; or one could hypothesise that firms might be able to efficiently contract, assigning the financing to the partner with the lowest cost of capital, which suggests that t ij (1 + min(r E, r I )) might be the appropriate measure. This could also be included in a horse race along side the prevailing capital costs in each trade partner, as a test of the efficient contracting hypothesis Trade Under Risk of Default The literature on endogenous contracting rightly emphasises the role contracts have in managing the real risks of non-performance, either by the importer, who might fail to pay on delivering the goods on open account terms, or by an exporter who might fail to deliver goods paid for in advance. Both Schmidt-Eisenlohr (2012) and Antràs & Foley (2013) emphasise the role that the quality of institutions, such as reputation for contract enforcement might have on shaping endogenous contracts: Antràs & Foley (2013) find that the probability of a US exporter bearing the financing risk rises with the quality of institutions in the importing country, and falls with the distance to market. However, this does not necessarily imply that the form of the contract is significant for trade volumes: in fact, if firms are vaying the terms of the trade credit they offer each other in line with the underlying default risk, then the party formally financing the contract may be irrelevant for final trade. If US exporters force customers in weak jurisdictions to finance their imports up front, and those importers face a high cost of finance because of large local financial frictions, then trade will be just as depressed as if the 19 Schmidt-Eisenlohr (2012) adopts a similar empirical strategy, except that he employs a double-log formulation, assuming the financial costs take the form ln(t ij ) ln(1 + r? ). 9

10 US exporter had agreed to finance the trade, pricing in the appropriate risk premium as a compensation for possible non-performance. Assume that the probability that a firm from country i defaults at time t on a contract signed at t = 0 (conditional on not having defaulted up to t), is prob(default at t i ) = λ i e λ it, where λ i is a country-specific default risk 20. so that the probability of default at some point before t is Prob(default by t) = 1 e λ it Assume a competitive international capital market, in which risk-free lenders receive the world risk-free rate, r W. Then lenders zero-profit condition implies that a term t loan will attract an interest rate of e λ it (1 + r i ) t = (1 + r W ) t and that the local cost of capital reflects both the world rate for the cost of time, and a local risk premium: ln(1 + r i ) = ln(1 + r W ) + λ i Returning to the problem of trade finance, again under the assumption of risk-neutral firms, an exporter who agrees to offer open account terms still faces marginal costs at the time of settlement of (1 + r i ) t ij τijc g i. However, his expected payment is now only e λit p E ij. His profit-maximisation problem becomes max(e λjt p ij c ij )x ij p ij which with CES demand implies p E ij = e λ σ jt 1 σ c ij = e λ σ jt 1 σ (1 + r i) t ij τijc g i = e (λ i+λ j )t (1 + r W ) t σ ij 1 σ τ ijc g i The price offered by an exporter on open account terms reflects all three components of both parties costs of capital: the world cost of time, and both risk-premia, through the exporter s own cost of capital, and the implicit risk premium he charges against the importer s default risk. A risk-neutral, competitive import distributor in j who finances his imports paying cash-in-advance, pays up-front p ij = σ τ g 1 σ ijc i to an exporter from i, which implies a certain cost by time of settlement of (1+r j ) t ij per unit imported. σ 1 σ τ g ijc i 20 This could be driven by the risk of a country-specific economic crises, such as a sharp change in exchange rates that would render a firm unable to pay a supplier, but is left unmodelled. 10

11 However, with probability 1 e λ it ij the exporter defaults and fails to deliver the goods at the time of settlement, so the import distributor must charge his customers a mark-up as an insurance against this default risk. The importer will charge final customers p I ij. His expected revenues are e λ it ij p I ijx(p I ij) The breakeven local price p I ij that equates total costs and revenues satisfies (1 + r j ) t ij σ 1 σ τ g ijc i x(p I ij) = e λ it ij p I ijx(p I ij) which implies a price to final consumers of p I ij = e λ it ij (1 + r j ) t ij σ 1 σ τ g ijc i = e (λ i+λ j )t ij (1 + r W ) t ij σ 1 σ τ g ijc i This is identical to the price consumers face when transactions were financed by the exporter, as it contains the same three terms: the world cost of time, the local risk premium, as embodied in the financing costs of the local importer, and the foreign risk premium, as compensation for the exporter s potential default risk. In a setting in which local costs of capital reflect risk premia, and participants in international trade set prices that take into account these default risks, the details of how trade is endogenously financed turn out to be irrelevant for the prices final consumers face, and hence their demand. The exposure to underlying default risks due to delayed settlement, and opportunity cost of having goods spend time in transit are unavoidable features of international transactions. The three capital cost terms function isomorphically to other trade barriers, and can be estimated along with them in a standard gravity equation. 3 Empirical Specification 3.1 Data The study analyses two datasets: one of aggregate cross-country bilateral trade at annual frequency for 172 countries from 1990 to 2006; the second of commodity-level imports to 26 destinations in the US at monthly frequency from 129 trade partners between January 1990 and December

12 3.1.1 Trade Flows The aggregate trade flows are taken from the UN Comtrade database, where differing reported trade flows are reconciled by using the importer s report if available, and otherwise adopting the exporter s declaration. US import data available from the US Census Bureau is highly disaggregated, across digit HS codes. For each commodity, the import data report the customs area in which the goods entered the US, and break down the value of imports that arrived by boat or by plane (the remainder coming by truck or rail from Mexico or Canada). There is enormous variation in transportation used across commodities and routes. An important challenge when working with panel data is to ensure the consistency of the commodities represented by the HS codes. This is nontrivial, as US customs periodically redefines its codes to reflect innovations in products, withdrawing and introducing new codes, frequently merging old codes together, or splitting an old code into several new ones as products evolve, and sometimes reintroducing an obsolete code for a completely different product. Fortunately Pierce & Schott (2010) have worked through the documentation of all the transitions and pieced together a concordance that links the connected HS codes over time, so that all commodities that are at any point defined within the same HS code, even if previously or subsequently they had distinct identities, are relabeled with the same identifier. I adopt their concordance technique, subject to a handful of revisions for cases where I was able to detect a typo in the original concordance 21. This distinguishes commodities that can be considered distinct families of products over time in this sample Interest Rates For both datasets, real interest rates are constructed by deflating nominal interest rates by the rate of CPI inflation, both taken from the IMF s International Financial Statistics database. When available, a country s Lending Rate is used, which is the average rate at which select banks lend to credit- 21 Preparing Pierce & Schott s (2010) concordance required matching tens of thousands of 10 digit numbers as their definitions change in the customs manuals, so unsurprisingly occasional typos slip in. These are relatively easily detected when a code s number of digits changes from 10 to 11 or 9. A typo in the first couple of digits of a code, that leads the commodity to dramatically changes goods categories, say from textiles to machinery, is also relatively easy to identify, and by checking the customs manuals to correct. Typos in later digits are much harder to distinguish from the actual transitions the concordance is trying to establish, and some may remain. 12

13 worthy business customers on 30 days loans, denominated in local currency. If the lending rate is not available, the country s Moneymarket Rate is substituted; if the Moneymarket Rate is also missing, the Discount Rate is used. For the first dataset, average interest rates and inflation prevailing over the calendar year are used. For the second dataset of US imports, monthly rates are constructed, and the geometric average of the current and previous monthly rates are used (ie the relevant interest rate for US imports arriving in January 1990 is the average rate applying between December 1989 and January 1990, as the decision to export may have been taken in the calendar month before the goods arrive. The unweighted average of shipping times between all pairs of ports in the dataset is 18.4 days). The first model of risk-free but segmented national capital markets calls for measures of both partners real interest rates, as well as the minimum of the two to test for the possibility of efficient contracting, yielding a specification of ln(τ f ij) = β E t ij ln(1 + r E ) + β I t ij ln(1 + r I ) + β min t ij ln(1 + min(r E, r I )) The second model, of integrated but risky capital markets requires information on the risk-premia, which are unobserved, but which can be proxied by local real rates, accounting for the duplication in the world risk free rate that this implies ln(τ f ij) = β E t ij ln(1 + r E ) + β I t ij ln(1 + r I ) β W t ij ln(1 + r W ) r W is not observed either, but it only has time, not cross-sectional variation, so it can be controlled for by estimating time-varying coefficients, β W,t, on the level of bilateral time-to-market, t ij Distance and Transit Times Average transit time between countries might reasonably be assumed to be proportional to the distance between them, which suggests t ij = αd ij can be substituted into (1) in an empirical implementation. Since geographical distance also matters directly for transportation costs, to avoid omitted variable bias that might arise from non-linearities in the model of τ ij, both the level and the log of distance are included as trade barriers, as well as the interaction terms d ij ln(1 + r X ). The measure of physical distance for the first dataset used is the great circle distance between countries largest cities. A drawback of using aggregate trade data is that it pools together goods that travel by ship with commodities transported by plane, road, or rail. While sea voyages between some trading partners are lengthy enough that 13

14 the transit time might be priced in 22, for goods transported by plane the difference in transaction times between domestic and international sales is likely negligible. The mode of transport used is partly determined by the technological characteristics of the commodity (e.g. weight-to-value ratio), that are unrelated to financial or macroeconomic conditions. Commodity-level trade data can provide additional variation that can be used to test the significance of working capital in trade, by testing whether tilting effects are observed for goods sent by sea, but not by plane. The analysis of the disaggregated commodity level data exploits the variation in transit times between 26 different points of entry into the US and 129 trade partners. All the entry points and trade partners for which data on direct shipping routes was available 23 were included in the analysis, and are listed in Tables 1 and 2, and illustrated in Figures 5 and 6. Only countries with a seaport in their own territory, for which direct shipping routes could be found, were included, as to include landlocked trade partners would require speculation over which neighbouring ports might be used, as well as potentially further adjustments to reflect transit times to third-country ports. When a trade partner had multiple active ports, the one with the largest volume of container shipping in 2010 was chosen as representative. Canada and Mexico are also dropped from the data, as most of their exports enter the US by road or rail. The disaggregated data distinguishes 43 areas of entry into the US, of which 37 are coastally located. Point-to-point shipping times and distances were available for 26 of these 37 ocean-going ports. The distance taken by plane is calculated as the great circle distance between an exporter s largest city and the US port. The continental scale of the US provides ample variation in the time it takes to ship goods from, e.g. Shanghai, to the east vs west coast, or to the Great Lakes (15 days to Seattle, 31 to New York, 37 to Milwaukee). There is also variation in the distance between ports, depending on the mode of transport: the great circle distance, the shortest path for a plane to take, between Shanghai and Milwaukee is km, while the distance travelled by ship is km. Using the rich pattern of commodity flows, we can explore whether trade-tilting effects appear for goods that travel by boat, but not by plane. 22 A ship travelling at 14 knots would take 31 days to cross the 19,155 km from Shanghai to New York ( 23 From which reports how long it would take a ship travelling at 14 knots to sail between two ports. It does not take into account how frequently such ships might actually be chartered, which would be endogenous to the expected volume of trade. 14

15 3.1.4 Other Trade Barriers A standard set of other trade barriers is included in the cross-country dataset: an indicator for whether countries share a common border; an indicator for whether both countries are landlocked, or both islands; an indicator for colonial ties, and for similar legal origins (following the classification of La Porta, de Silanes, Shleifer & Vishny (1998)); an indicator for sharing a common currency or a fixed exchange rate regime (following the classification of Ilzetzki, Reinhart & Rogoff (2008)); and an index of religious similarity, constructed as a Herfindahl index of local religious denominations, using data from Barrett, Kurian & Johnson (2001). No attempt is made to model regional variation in institutions across the US, so in the dataset of US imports all the variation associated with these variables will be collinear with exporter dummies, and they need not be explicitly modelled. 3.2 Evidence from aggregate trade flows Table 3 presents estimates of a standard cross-country gravity equation, including the interactions between distance, proxying for shipping time, and the cost of capital. The specification in column (1) includes only the trade barriers, while column (2) adds time dummies, column (3) includes dummies for the importer and exporter, and column (4) has separate importer and exporter dummies for each year in the panel. A clear pattern of trade tilting away from long-haul trade routes by partners facing high interest rates emerges across all specifications, as evidenced by the significant negative coefficients estimated on the interactions of both the importer s and exporter s interest rate with distance. These results differ from Schmidt-Eisenlohr (2012), who found that only the minimum of the trade partners interest rates had a statistically significant negative effect, which could be interpreted as evidence in favour of flexible contracting, allocating the burden of financing to the firm with access to the cheapest capital. The results in Table 3 are not supportive of this result, as the interaction of distance and the minimum interest rate is positive, while the interaction with both the importer s and exporter s cost of capital retain their negative significance, suggesting that the financing burden is not easily shifted 24. They also differ from Paravisini et al. (2012), who include an interaction of credit and distance in some specifications but found no effects. 24 Schmidt-Eisenlohr (2012) uses a log-log interaction, which is less easily mapped back into a model of financing costs, as well as nominal interest rates. 15

16 In specifications (1)-(3) country-specific time-varying variables can be estimated, and the level of interest rates appears to correlate strongly with bilateral trade, with high interest rates discouraging exports and encouraging imports. One obvious mechanism through which this could be arising is through movements in the real exchange rate, a separate channel to the one of principal interest here. This illustrates the importance of robustly controlling for country-level macro factors through time-varying importer and exporter dummies, as in specification (4), as well as identifying the importance of working capital off of the tilting mechanism, given by the interaction terms. One possible concern when analysing a large cross-country dataset is that the quality of information on local financial conditions might be very variable. In particular, a measure of the cost of capital could be misleading in environments in which extensive credit rationing is widespread. Since this is likely to be more of a concern in developing than developed economies, Table 4 repeats the analysis with time-varying importer/exporter dummies for increasingly restrictive subsets of the data. The first column shows restricts one of the trade partners to be an OECD member; column (2) includes only exports of OECD members, to all trade partners; column (3) imports of OECD countries from all trade partners; and column (4) looks only at trade between two OECD members. The pattern of results is consistent across these different subsets of trade flows: the tilting terms of both importer and exporter appear to be important, while the minimum of the two rates actually has a positive estimated coefficient, which does not support the hypothesis of flexible financing. Interpreting the magnitude of these results, a coefficient on the interaction term of -0.3 implies that an additional 1000km of separation increases the elasticity of trade with respect to interest rates by 0.3%. The mean distance for trade in this data is 7,500km, so at this distance, a 1% increase in real interest rates would lead to a 2.25% fall in trade. US real interest rates increased by 3.24% between June 2008 and January 2009, and the average (unweighted) distance to US trading partners is 8000km. A back of the envelope calculation, relating these estimates to the trade collapse of the Great Recession, suggests that US imports and exports with the average trade partner would have fallen by 7.8%, in addition to any shocks driven by changes to GNI, as a result of that increase in the cost of working capital. Total US exports fell 10.2% more than GDP from Q2: Q4:2008, while US imports fell 12.8% faster than GNI. 16

17 3.3 Evidence from disaggregate US imports To investigate further whether a higher cost of working capital has a discouraging effect on trade, we can compare effects across deliveries sent by plane, which are effectively as rapid as domestic transactions, and consignments sent by ship, for which the cost of time might be expected to matter. A simple estimating equation can be derived from a basic model of CES demand for products differentiated by origin i and by type l. Region j has utility function ( 1/α u j = x j (i, l) dldi) α (2) where ɛ = 1/(1 α) is the elasticity of substitution across varieties x(i, l). Local demand for variety (i, l) is x j (i, l) = p j(i, l) ɛ Y j P 1 ɛ j (3) where ( P j = p j (i, l) 1 ɛ dldi) 1/(1 ɛ) (4) is the aggregate price index in region j and the optimal price of variety l, produced in country i and sold in region j, is a mark-up over the unit cost of production and transportation, p j (i, l) = τ ijc i (l) α (5) Assume that two varieties l and l are sufficiently similar that they are classed as being within the same HS code, and have identical unit costs of production, c i (l) = c i ( l) if produced in the same country i; but they differ in their transportation cost, τ and τ, in that l is transported by boat and l by plane. ln x j (i, l) = ln Y j (1 ɛ) ln P j + ɛ ln α ɛ ln c i (l) ɛ ln τ ij (6) ln x j (i, l) = ln Y j (1 ɛ) ln P j + ɛ ln α ɛ ln c i ( l) ɛ ln τ ij (7) Then for this pair of goods, similar enough to be classified within the same HS category, ln x j (i, l) ln x j (i, l) = ɛ ln τ ij ɛ ln τ ij (8) 17

18 and the difference in the value shipped versus flown along the same route can be used to estimate differences in the cost of the modes of transportation along that route, differencing out demand and supply conditions in the purchasing and producing regions respectively. Taking equation (8) to the disaggregated US import data, there are over 3 million instances in which the same commodity was both shipped and flown to the same point of entry in the US from the same exporting country in a particular month. Air transportation costs τ are assumed proportional to the great circle distance between origin and destination. Following equation (1), shipping transportation costs τ are assumed to depend on the shipping distance, as well as the time cost of shipping, which involves the interaction of time with the log interest rate. Table 5 reports estimates of equation (8). As above, the number of days to ship is also included directly to ensure that the interaction term is not loading on any non-linearities in the physical costs of transportation. Columns (1)- (4) vary the combinations of the interaction terms. In all cases, the direct transportation costs appear with the signs one would expect: ceteris paribus, routes requiring relatively roundabout shipping paths see significantly lower volumes of trade travelling by boat relative to plane, and vice versa. The continental scale of the US introduces sufficient variation in the distance between two points by boat and plane to identify these effects very strongly. It is also the case that the time cost of shipping appears to significantly reduce the value of goods sent by boat. Similarly to the aggregate data, columns (1)-(3) suggest that interest rates in both the importer and exporter may be important. However column (4) suggests that unlike in the aggregate data, the minimum interest rate may be used to achieve efficient financing, as the value of goods sent by boat falls over long distances as the minimum interest rate rises. In column (4) the US interest rates retains its negative and significant effect, while the exporting country s interest rate appears positive. This difference from the results on the aggregate dataset may reflect something special about trade with the US, as with its more sophisticated financial markets, it may be easier to make flexible financing contracts to take advantage of the lowest available interest rates, and this might not be a general feature of trade. To compare the magnitude of the estimates across the two datasets, adjustment should be made for the fact that in Tables 3 and 4 distance was used as a proxy for shipping times, while in Table 5 the number of days is used directly. The distance travelled by a ship in one day is approximately 622 km 25, so to convert the interaction terms into similar units, the coeffi- 25 Assuming a speed of 14 knots. 18

19 cients in Table 5 should be scaled up by a factor of 1000/ , which brings them closer towards those estimated on the aggregate data, although still somewhat smaller. 4 Conclusions Trade does appear to tilt away from long-haul destinations when the cost of capital rises. Identifying the importance of financial factors from a tilting effect allows one to control for aggregate macroeconomic shocks which might otherwise be endogenously co-determined with trade flows and financial conditions. The finding that the tilting phenomenon differentially affects goods travelling by boat significantly more than goods that go by plane confirms that the transit time is the critical mechanism at work. Although not estimated from data drawn from the recent financial crisis, the magnitude of the estimates suggests that the sharp increase in real interest rates as financial and macroeconomic conditions deteriorated in the last quarter of 2008 and beginning of 2009 could have played a quantitatively significant part in the collapse of global trade flows, and that the efforts of central banks around the world to successfully bring real interest rates down in the aftermath of the crisis may have played an equally important role in trade s subsequent strong recovery. References Ahn, J. (2011). A theory of domestic and international trade finance. Ahn, J., Amiti, M. & Weinstein, D. E. (2011). Trade finance and the great trade collapse, American Economic Review 101(3): Alessandria, G., Kaboski, J. P. & Midrigan, V. (2010). The great trade collapse of : An inventory adjustment? NBER Working Paper Amiti, M. & Weinstein, D. E. (2011). Exports and financial shocks, Quarterly Journal of Economics 126(4): Anderson, J. E. & van Wincoop, E. (2003). Gravity with gravitas: A solution to the border puzzle, 93(1): Antràs, P. & Foley, C. F. (2013). Poultry in motion: A study of international trade finance practices. 19

20 Auboin, M. (2009). Restoring trade finance during a period of financial crisis: Taking stock of recent initiatives. World Trade Organisation Staff Working Paper ERSD Baldwin, R. (ed.) (2009). The Great Trade Collapse: Causes, Consequences and Prospects, VoxEU.org. Barrett, D. B., Kurian, G. T. & Johnson, T. M. (2001). World Christian Encyclopedia, Oxford University Press. Behrens, K., Corcos, G. & Mion, G. (2010). Trade crisis? what trade crisis? National Bank of Belgium Working Paper no Bems, R., Johnson, R. C. & Yi, K.-M. (2010). Demand spillovers and the collapse of trade in the global recession. IMF Working Paper 10/142. Berman, N., De Sousa, J., Martin, P. & Mayer, T. (2012). Time to ship during financial crises. NBER Working Paper Berman, N. & Héricourt, J. (2008). Financial factors and the margins of trade: Evidence from cross-country firm-level data. CES Working Paper. Bricongne, J.-C., Fontagné, L., Gaulier, G., Taglioni, D. & Vicard, V. (2012). Firms and the global crisis: French exports in the turmoil, Journal of International Economics 87(1): Chor, D. & Manova, K. (2011). Off the cliff and back? credit conditions and international trade during the global financial crisis. Djankov, S., Freund, C. & Pham, C. S. (2010). Trading on time, Review of Economics and Statistics 92(1): Eaton, J., Kortum, S., Neiman, B. & Romalis, J. (2011). global recession. NBER Working Paper Trade and the Feenstra, R. C. (2004). Advanced International Trade: Theory and Evidence, Princeton University Press. Feenstra, R. C., Li, Z. & Yu, M. (2011). Exports and credit constraints under incomplete information: Theory and evidence from china. NBER Working Paper FImetrix (2009). IMF-BAFT Trade Finance Survey: A Survey Among Banks Assessing the Current Trade Finance Environment. 20

21 Freund, C. (2009). The trade response to global downturns: Historical evidence. World Bank Policy Research Working Paper no Hummels, D. & Schaur, G. (2012). Time as a trade barrier. Iacovone, L. & Zavacka, V. (2009). Banking crises and exports: Lessons from the past. World Bank Policy Research Working Paper no Ilzetzki, E., Reinhart, C. M. & Rogoff, K. S. (2008). The country chronologies and background material to exchange rate arrangements in the 21st century: Which anchor will hold? La Porta, R., de Silanes, F. L., Shleifer, A. & Vishny, R. W. (1998). Law and finance, Journal of Political Economy 106(6): Levchenko, A. A., Lewis, L. T. & Tesar, L. L. (2010a). The collapse of international trade during the crisis: In search of the smoking gun, IMF Economic Review 58(2): Levchenko, A. A., Lewis, L. T. & Tesar, L. L. (2010b). The role of financial factors in the trade collapse: A skeptic s view. Muûls, M. (2012). Exporters, importers and credit constraints. CEP Discussion Paper Paravisini, D., Rappoport, V., Schnabl, P. & Wolfenzon, D. (2012). Dissecting the effect of credit supply on trade: Evidence from matched credit-export data. Pierce, J. R. & Schott, P. K. (2010). Concording us harmonized system codes over time. Ronci, M. (2004). Trade finance and trade flows: Panel data evidence from 10 crises. IMF Working Paper 225. Schmidt-Eisenlohr, T. (2012). Towards a theory of trade finance. Taylor, A. M. & Taylor, M. P. (2004). The purchasing power parity debate, Journal of Economic Perspectives 18(4): WTO (2009). International Trade Statistics. WTO (2010). International Trade Statistics. WTO (2011). International Trade Statistics. 21

22 Table 1: Ports of Entry into the United States Boston, MA New York, NY Baltimore, MD Charleston, SC Mobile, AL Port Arthur, TX Los Angeles, CA San Diego, CA Seattle, WA Milwaukee, WI Chicago, IL Newark, NJ San Juan, PR Buffalo, NY Philadelphia, PA Norfolk, VA Tampa, FL New Orleans, LA Houston-Galveston, TX San Francisco, CA Duluth, MN Honolulu, HI Detroit, MI Cleveland, OH Miami, FL US Virgin Islands Table 2: Ports of US Trading Partners Algiers, Algeria St. Johns, Antigua Oranjestad, Aruba Nassau, Bahamas Chittagong, Bangladesh Antwerp, Belgium Santos, Brazil Varna, Bulgaria San Antonio, Chile Macao, Macao Cartagena, Colombia Matadi, DR Congo Caldera, Costa Rica Rijeka, Croatia Copenhagen, Denmark Roseau, Dominica Guayaquil, Ecuador Acajutla, El Salvador Tallinn, Estonia Kotka, Finland Libreville, Gabon Luanda, Angola Buenos Aires, Argentina Melbourne, Australia Mina Sulman, Bahrain Bridgetown, Barbados Cotonou, Benin Muara, Brunei Douala, Cameroon Hong Kong, Hong Kong Shanghai, China Moroni, Comoros Pointe-Noir, Rep Congo Abidjan, Côte d Ivoire Limassol, Cyprus Djibouti, Djibouti Haina, Dominican Republic Alexandria, Egypt Bata, Equatorial Guinea Suva, Fiji Marseille, France Banjul, Gambia 22

23 Table 2: Ports of US Trading Partners Poti, Georgia Tema, Ghana Santo Tomas, Guatemala Bissau, Guinea-Bissau Port-au-Prince, Haiti Reykjavik, Iceland Tanjung Priok, Indonesia Haifa, Israel Kingston, Jamaica Aqaba, Jordan Busan, Rep of Korea Riga, Latvia Tripoli, Libya Toamasina, Madagascar Male, Maldives Nouakchott, Mauritania Plymouth, Montserrat Maputo, Mozambique Walvis, Namibia Willemstad, Netherlands Antilles Corinto, Nicaragua Oslo, Norway Karachi, Pakistan Port Moresby, Papua New Guinea Callao, Peru Gdansk, Poland Doha, Qatar St Petersburg, Russia São Tomé, São Tomé & Principe Dakar, Senegal Port Victoria, Seychelles Singapore, Singapore Honiara, Solomon Islands Algeciras, Spain Basseterre, St Kitts & Nevis Kingstown, St Vincent Gotheburg, Sweden Dar es Salaam, Tanzania Lome, Togo Hamburg, Germany Piraeus, Greece Conakry, Guinea Georgetown, Guyana Puerto Cortes, Honduras Mumbai, India Dublin, Ireland Gioia Tauro, Italy Tokyo, Japan Mombasa, Kenya Shuaiba, Kuwait Monrovia, Liberia Klaipeda, Lithuania Port Kelang, Malaysia Marsaxlokk, Malta Port Louis, Mauritius Tangiers, Morocco Rangoon, Myanmar Rotterdam, Netherlands Auckland, New Zealand Apapa, Nigeria Muscat, Oman Balboa, Panama Asuncion, Paraguay Manila, Philippines Lisboa, Portugal Constanta, Romania Apia, Samoa Jeddah, Saudi Arabia Bar, Serbia Freetown, Sierra Leone Koper, Slovenia Durban, South Africa Colombo, Sri Lanka Vieux Fort, St Lucia Paramaribo, Suriname Latakia, Syria Laem Chabanga, Thailand Port of Spain, Trinidad & Tobago 23

24 Table 2: Ports of US Trading Partners Tunis, Tunisia Odessa, Ukraine Montevideo, Uruguay La Guaira, Venezuela Aden, Yemen Mersin, Turkey Felixstowe, United Kingdom Santo, Vanuatu Ho Chi Minh City, Vietnam % exports/gdp imports/gdp FFR Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Figure 1: US Trade and nominal Federal Funds Rate 24

25 dist*ln(int E ) ** ** ** ** (0.0385) (0.0370) (0.0334) (0.0335) dist*ln(int I ) ** ** ** ** (0.0401) (0.0385) (0.0348) (0.0345) dist*ln(min(int E, int I )) 0.500** 0.682** 0.665** 0.416** (0.0541) (0.0519) (0.0469) (0.0392) distance (1000 km) ** ** ( ) ( ) ( ) ( ) ln(distance) ** ** ** ** (0.0170) (0.0163) (0.0162) (0.0163) ln(int E ) ** ** 1.337** (0.330) (0.323) (0.301) ln(int I ) 1.053** 0.736* 2.500** (0.340) (0.333) (0.310) ln(min(int E, int I )) ** ** (0.461) (0.448) (0.418) ln(gdp I ) 0.946** 0.953** 0.808** ( ) ( ) (0.0272) ln(gdp E ) 1.136** 1.143** 0.570** ( ) ( ) (0.0272) border ** ** (0.0412) (0.0395) (0.0359) (0.0356) both islands 0.355** 0.351** ** ** (0.0134) (0.0129) (0.0275) (0.0273) both landlocked ** ** ** ** (0.0149) (0.0143) (0.0474) (0.0471) legal 0.243** 0.298** 0.375** 0.382** (0.0147) (0.0141) (0.0136) (0.0135) language 0.623** 0.458** 0.427** 0.407** (0.0178) (0.0172) (0.0183) (0.0183) colonial 0.784** 0.619** 0.616** 0.607** (0.0306) (0.0294) (0.0279) (0.0277) religion ** ** 0.581** 0.593** (0.0285) (0.0273) (0.0279) (0.0278) common currency 0.967** 1.247** 1.503** 1.464** (0.0453) (0.0436) (0.0413) (0.0432) fixed exchange rate ** ** 0.313** 0.343** (0.0252) (0.0244) (0.0235) (0.0250) Observations 110, , , ,337 R-squared Dummies Year n y y Importer/Exporter n 25 n y Importer/Exporter-Year n n n y Standard errors in parentheses ** p<0.01, * p<0.05 Table 3: Tilting of Cross-Country Aggregate Bilateral Trade

26 All OECD OECD OECD intra-oecd exports & exports imports trade imports only only only ln(distance) ** ** ** ** (0.0177) (0.0173) (0.0237) (0.0186) distance (1000 km) ** ** ** ** ( ) ( ) (0.0051) ( ) dist*ln(int E ) ** ** * ** (0.0391) (0.0417) (0.053) (0.0499) dist*ln(int I ) ** ** ** ** (0.0394) (0.0389) (0.0577) (0.0499) dist*ln(min(int E, int I )) 0.573** 0.513** 0.621** 0.510** (0.0448) (0.0456) (0.0632) (0.059) border ** ** (0.0378) (0.0326) (0.0443) (0.0265) both islands ** ** ** ** (0.0269) (0.0268) (0.0370) (0.0307) both landlocked * ** * ** (0.0415) (0.0417) (0.0578) (0.0507) legal 0.480** 0.399** 0.530** 0.343** (0.0137) (0.0136) (0.0186) (0.0151) language 0.242** 0.210** 0.148** ** (0.0182) (0.0178) (0.0243) (0.0185) colonial 0.652** 0.449** 0.628** 0.160** (0.0227) (0.0221) (0.0301) (0.0225) religion 0.265** 0.383** 0.168** 0.388** (0.0301) (0.0289) (0.0398) (0.0291) common currency ** * (0.0503) (0.0461) (0.0627) (0.0425) fixed exchange rate 0.406** 0.382** 0.400** 0.149** (0.0221) (0.0223) (0.0304) (0.0269) Observations 74,078 46,749 45,629 18,300 R-squared Dummies Importer/Exporter-Year y y y y Standard errors in parentheses ** p<0.01, * p<0.05 Table 4: Tilting of Aggregate Bilateral Trade of OECD Trade Partners 26

27 ln(ship dist) ** ** ** ** (0.0158) (0.0157) (0.0158) (0.0158) ln(air dist) 3.151** 3.102** 3.149** 3.157** (0.0102) (0.0101) (0.0102) (0.0102) days to ship ** ** ** ** ( ) ( ) ( ) ( ) days*ln(int E ) ** ** ** ( ) ( ) ( ) days*ln(int US ) ** ** ** ( ) ( ) ( ) days*ln(min(int E, int US )) ** (0.0017) Observations 3,319,567 3,325,445 3,319,567 3,319,567 R-squared Standard errors in parentheses ** p<0.01, * p<0.05 Table 5: Route-Specific Difference in Shipped vs Flown US Imports BoJ money call rate exports/gdp imports/gdp % Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Figure 2: Japanese Trade and nominal Bank of Japan call rate 27

28 exports/gdp imports/gdp real interest rate % Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Figure 3: US Trade and real Federal Funds Rate Real corporate lending rate exports/gdp imports/gdp % Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Figure 4: Japanese Trade and real Bank of Japan call rate 28

29 Figure 5: Entry Ports into the US Figure 6: Principal Ports of US Trade Partners 29