WPS6226 Policy Research Working Paper 6226 Innocent Bystanders How Foreign Uncertainty Shocks Harm Exporters Daria Taglioni Veronika Zavacka The World Bank Poverty Reduction and Economic Management Network International Trade Department October 2012 Policy Research Working Paper 6226 Abstract The failure of trade economists to anticipate the extreme The paper does not find evidence of learning from drop in trade post Lehman Brothers bankruptcy suggests past turmoils, suggesting that prior experience with that the behavior of trade in exceptional circumstances major uncertainty shocks does not reduce the effect may still be poorly understood. This paper explores on trade. In line with the expectations, the negative whether uncertainty shocks have explanatory power effect of uncertainty shocks on trade is higher for for movements in trade. VAR estimations on United trade relationships more intensive in durable goods. States data suggest that domestic uncertainty is a strong Surprisingly, however, the effect of durability is non- predictor of movements in imports, but has little effect linear. Supply chain considerations or the possibility on exports. Guided by these results, the paper estimates that the relationships with the highest durability lead to a bilateral model with focus on the impact of importer important compositional effects may have a bearing on uncertainty on foreign suppliers. It finds that there is the results. The results are robust to excluding the post a strong negative relationship between uncertainty and Lehman shock, suggesting that the trade response during trade and that this relationship is non-linear. Uncertainty the 2008–2009 crisis has been similar to past uncertainty matters most when its levels are exceptionally high. events. This paper is a product of the International Trade Department, Poverty Reduction and Economic Management Network. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank. org. The authors may be contacted at dtaglioni@worldbank.org or veronika.zavacka@graduateinstitute,ch. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Innocent Bystanders: ∗ How Foreign Uncertainty Shocks Harm Exporters Daria Taglioni† Veronika Zavacka‡ October 1, 2012 JEL Classi�cation: F02, F10, G01 Keywords: uncertainty shocks, international trade, exporters Sector Board: Economic Policy (EPOL) ∗ This paper represents the views of the authors and should not be interpreted as reflecting those of the European Bank for Reconstruction and Development or the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. We acknowledge valuable comments by George Alessandria, Jean-Louis Arcand, Richard Baldwin, Nicolas Berman, Matthieu Bussiere, Giancarlo Corsetti, Thomas Farole, Bernard Hoekman, Daniel Lederman and Ugo Panizza. † The World Bank. Email: dtaglioni@worldbank.org ‡ Graduate Institute, Geneva and European Bank for Reconstruction and Development. Email: veronika.zavacka@graduateinstitute.ch, zavackav@ebrd.com 1 1 Introduction The exceptional features of the 2008-2009 trade crisis were largely unanticipated by trade economists.1 Simulations which focused on the 2008–2009 trade crisis, aimed at identifying the contribution of demand and that controlled for international input-output relationships, have hardly reproduced the magnitude of the slump in world exports (Benassy-Quere et al., 2009; Willenbockel and Robinson, 2009). Similarly, standard trade forecasting models, which only account for �real-economy� mechanisms, show a sudden increase in the unexplained residual in coincidence with the recent crisis (Cheung and Guichard, 2009; Levchenko et al., 2010).2 The failure of traditional models to account for these developments suggests that the behavior of trade in exceptional circumstances may still be poorly understood. Some economists have proposed new structural measures of aggregate demand for empirical trade equations (Bussiere et al., 2011). The current paper, instead, investigates how uncertainty and con�dence factors affect international trade. Understanding this nexus may contribute to shed light on some of the potentially harmful effects of globalization (Rodrik, 1997, 2011). In particular, we argue that while international trade developments are usually well explained by demand and cross-country differences in competitiveness, under exceptional circumstances con�dence factors may also have a large impact on trade flows. This is consistent with the real-business cycle literature according to which a temporary increase in uncertainty can have non- negligible effects on aggregate activity. Why can one expect extreme movement and unusually high variability in �nancial markets to generate temporary uncertainty about future incomes, even among people that do not hold �nancial assets? Financial markets can be considered an imperfect but readily available predictor of the real economy. Hence, it is likely that consumers and producers alike associate unusual developments in �nancial markets with greater uncertainty as to the future of the economy. Casual observation suggests indeed that con�dence and credit crises are likely to happen jointly, as they rely on the same collective predictions as to the state of the economy. After all, their Latin roots indicate that they reflect similar sentiments: the word �con�dence� comes from the Latin �do, which means �I trust, I believe� while the word �credit� comes from the latin credo, which also means �I trust, I believe�. To date the analysis of the effects of uncertainty on trade remains largely con�ned to exchange rate volatility, where studies broadly �nd that the effect of exchange rate volatility on aggregate trade flows is �fairly small and by no means robust� (IMF, 2004). Yet, a crisis of con�dence may easily 1 In the last quarter of 2008 and in the �rst quarter of 2009 trade contracted in an exceptionally sudden, severe and globally synchronized fashion. This great trade collapse was unparalleled in its suddenness: the decline of world trade totalled, in value, 29% in just four months, from September 2008 to January 2009. It was also seemingly out of line with the decline of world GDP, which only contracted by less than 3% over the same period. 2 The two main determinants of a country’s exports accounted by forecasting models are foreign demand and its com- petitiveness position relative to other countries competing on foreign markets. Foreign demand is usually computed as a weighted average of the import volumes of trading partners. Export competitiveness is measured as the relative price between domestic prices and foreign prices, both measured in a common currency. Similarly, the two main determinants of a country’s imports accounted for by forecasting models are domestic expenditure and its price-competitiveness relative to other countries competing on the domestic market. In both cases, under normal conditions empirical estimates attribute to demand about 70-75% of the overall trade changes and to relative prices another 20-25%. In such speci�cations, purely based on production and demand forces, exchange rate developments enter as a component of relative prices. Financial conditions, on their part, enter the function indirectly, as a �structural� element that contributes to explain countries trade potential and geographical orientation. Uncertainty has no explicit role and enters, at best, through fluctuations in demand. 2 curtail international trade, as investment and consumption decisions are put on hold. Investment and consumption as well as money markets and banks’ willingness and capacity to lend have been shown to be reduced in the presence of a crisis of con�dence. For example, during the Great Depression uncertainty played a role in the initial 1929-1930 slump, which was propagated - also internationally - by the 1931 banking collapse (Romer, 1990).The temporary increase in uncertainty also caused an immediate drop in investment spending, as discussed in Bernanke (1983). Indeed, while TFP fell by 18% between 1929 and 1933 (Ohanian, 2002), output did not shift to low-cost �rms (Bresnahan and Raff, 1991), as it may have done if only real-economy mechanisms were at play. In this paper we assess the importance of exceptional systemic uncertainty in explaining trade devel- opments during and in the aftermath of a major shock. We believe that this question is very relevant to understanding recent trade developments. Bloom (2009) shows that uncertainty tends to increase dra- matically after major economic or political shocks. There is consensus that the global crisis represents indeed a major economic shock likely to have generated large negative con�dence effects.3 To the best of our knowledge, this is the �rst paper that tackles explicitly the behavior of trade in response to exceptionally high systemic uncertainty. We use a stock market volatility measure as a proxy for uncertainty, following a common practice in the uncertainty literature. Even though past literature has not been concerned with the impact of uncertainty on international trade, a lot has been written about its impact on producer and consumer behavior. Both strands of the uncertainty literature (i.e. the one addressing producer behavior and the one on consumer behavior) imply that, under uncertainty, economic agents freeze their activity and postpone decisions on purchases. This is the case, in particular, for those goods whose purchase is partially or totally irreversible for long periods of time. Capital goods, acquired by producers, and durable goods, acquired by consumers, fall into the category of goods whose purchase decision is likely to be postponed. The inaction persists until uncertainty subsides and it gives rise to pent-up demand in the medium term, so that after a period of uncertainty there might be an overshooting effect. This is indeed what happened in late 2008 and early 2009. After an extreme drop, for some countries representing as much as 30% of their total value of exports, trade returned on a relatively strong path to recovery. We investigate the uncertainty hypothesis �rst by means of a vector autoregression model in an analysis very similar to Bloom (2009). However, we expand his analysis, that is entirely focused on producer behavior, in two respects. First, we address consumer uncertainty and, second, we apply the framework to a setting of international trade. In contrast, Bloom models and simulates the joint effects of time-varying uncertainty on labor and capital investment for the �rm, abstracting from any international dimension. Our results show that uncertainty shocks disproportionately affect imports. Our next step is 3 The “Subprime Crisis� broke out in August 2007 and for over one year it was broadly viewed as a �nancial crisis restricted mainly to those few industrialised countries with �nancial markets developed enough to absorb large quantities of the sophisticated �nancial derivatives, which were at the origin of the crisis. The metastasis into the “Great Global Recession� took place in September 2008, when a rapid sequence of extreme events plunged the world into “Knightian uncertainty�, or fear of the unknown (Blanchard, 2009; Caballero, 2009a,b). Consumers, �rms, and investors around the world applied a strategy of “wait and see� by delaying investment and purchases of all what could be postponed. Investors massively switched their wealth to the safest assets, causing what Caballero has called a “sudden �nancial arrest�, leading to deleveraging and a retrenchment of investment, often towards domestic assets (Kamil and Rai, 2009). In 2012 the crisis is still ongoing with the economic side effects of the European sovereign debt crisis, accompanied with slow US and Chinese growth. 3 therefore to analyze empirically the effect of importer uncertainty on exports. We do so using a dataset of bilateral trade between 32 developed and developing countries. We �nd that elevated uncertainty has signi�cant negative effects on trade even when controlling for potentially confounding factors such as �nancial constraints and reductions in wealth, that tend to accompany major uncertainty periods. In line with our expectations, the negative effect of uncertainty shocks on trade is higher for trade relationships more intensive in durable goods. Surprisingly however, the relationship is non-linear: we �nd that the top trade relationships in durables intensity are resilient to uncertainty. Supply chain considerations or the possibility that the relationships with the highest durability lead to important compositional effects may have a bearing on the results. This �nding calls for additional research on the linkages between durability, uncertainty and trade. Lack of intra-annual bilateral data by product category however does not allow us to identify better the role played by durability. We also show that prior experience with major uncertainty shocks does not reduce the effect on trade, i.e., we do not �nd evidence of learning from prior shocks that would help smooth out the adverse impact. Finally, the response of trade to uncertainty in the 2008-2009 crisis reflected the behavior of trade in past con�dence crises. The difference was in the size of the drop and subsequent recovery which was much stronger in the most recent crisis compared to the past. The rest of the paper is organized as follows. Section 2 reviews the insights from the theory and discusses the predictions for the impact of uncertainty on aggregate trade. Section 3 provides a prelim- inary empirical analysis using the U.S. data. Section 4 discusses our methodology and data. Section 5 brings the empirical investigation to an international setting. By means of a dynamic bilateral model it investigates the impact of exceptional uncertainty on imports and tests the main predictions outlined in Section 2. Finally, Section 6 concludes and draws the implications of our results. 2 Producer and consumer uncertainty The fact that a temporary increase in uncertainty can cause an immediate drop in investment is discussed in Bernanke (1983) and extended to the effects of income uncertainty on consumer spending by Romer (1990). At the same time, Dixit (1989a) shows that, in making entry and exit decisions, uncertainty about future prices creates an option value of waiting until more information about the state of the world is received. The generality of the idea is perfected and shown to have many applications in Dixit and Pindyck (1994). Building on a body of literature that became very rich in the past two decades, a recent paper (Bloom, 2009) provides a structural framework to analyze the effect of uncertainty shocks jointly on investment and hiring. In the following sections we will summarize the key concepts from this body of research and explain how this applies to international trade. 2.1 Producer uncertainty Bernanke (1983) discusses how an increase in the cost of credit intermediation can account for the link between uncertainty and investment spending. Later research (Dixit, 1989a; Dixit and Pindyck, 1994) emphasizes that the changes in investment spending are the outcome of decisions that are very often 4 made in an uncertain environment and are costly to reverse later. More precisely, for producers facing increasing returns, the relative payoff to various investment projects will depend on the uncertain level of future income. Besides being subject to some degree of uncertainty over the future rewards, most investment decisions share two additional important characteristics. First, there is some leeway about the timing of the investment, i.e. the investor can postpone action to get more information about the future. Second, investment decisions are subject to non-negligible adjustment costs. The latter can be convex or non-convex.4 The above characteristics interact to determine the optimal decisions of investors. For the �rm, the optimization problem becomes one of maximizing the present discounted flow of net revenues while also accounting for the non-convex adjustment costs and for the non-linear nature of investment and hiring decisions. It follows that a wait-and-see attitude in investment and hiring may be a rational decision under certain circumstances. Firms only hire and invest when business conditions are sufficiently good, and only �re and disinvest when business conditions are sufficiently poor. When uncertainty increases, �rms become more cautious in responding to business and macroeconomic conditions. It follows that producers have an option value of waiting. The existence of values for which the real option value of waiting is worth more than the returns to investment, disinvestment, hiring or �ring gives rise to a region of inaction. Only outside the region of inaction decisions are governed by the discounted value of the respective marginal returns. The generality of the idea and the variety of its application is brought out by Dixit (1989a) and the joint effects of time-varying uncertainty on labor and capital investment for the �rm are modeled and simulated by Bloom (2009). All in all, one can expect that the option value of waiting for producers is increasing in uncertainty and in the amount of implied non-convex adjustment costs. 2.2 Consumer uncertainty While the more recent literature has mainly analyzed the reaction of �rms to uncertainty, there is a substantial body of literature showing that uncertainty also has strong effects on consumer behavior. Dixit (1989a) and Romer (1990) among others point out that the intuition of why uncertainty might depress investment spending can straightforwardly be applied to consumer spending. Under uncertainty, consumers might �nd it advantageous to delay the purchase of goods, in particular those whose purchase is partially or totally irreversible for long periods of time, i.e. durable goods. More precisely, not knowing the value of their future income, consumers may choose a consumption bundle that is either too luxurious or too modest relative to their future level of income. On the other hand, if they wait, they will be very far from the optimal level of consumption while waiting but then, as soon as the uncertainty about 4 While early studies on investment mainly discussed convex adjustment costs, i.e. those whose marginal cost is increasing in the rate of investment, more recently the literature on investment and employment adjustment costs typically focuses on three types of non-convex costs: irreversibilities, �xed disruption costs and quadratic adjustment costs. Irreversibilities indicate that labor and capital investment are totally or partially irreversible. labor partial irreversibilities are, for example, the per capita hiring, training, and �ring costs while examples of capital partial irreversibilities are instead the resale losses due to non-recoverable transaction costs, the market for lemons phenomenon, and the physical cost of resale. Fixed disruption costs, on their part, arise when a loss of output is generated due to the need to integrate new workers into the production process or to install new capital. Finally, quadratic adjustment costs are related to the rate of adjustment due to higher costs for more rapid changes. 5 future income is resolved, they will be able to choose the appropriate level of consumption. Hence, one can expect that, intertemporally, the option value of waiting for consumers is increasing in the weight of durable goods in the consumer expenditure function and it is also higher the longer the irreversibility of their purchase. The adjustment mechanism that governs the purchases of consumption goods is critical for under- standing why durables consumption in particular will overreact under uncertainty. In early work on this topic Bernanke (1984, 1985) uses a rule that implies a regular adjustment with which the consumers close the gap between their desired and actual levels of durables stock. However, this type of behavior is not consistent with the data in which adjustments happen only infrequently. This observation has led subsequent studies to address the shortcoming of Bernanke by modelling the adjustment mechanism via the (S,s) model framework �rst used by Arrow et al. (1951) in the context of inventories adjustment, e.g. Lam (1991); Attanasio (2000); Eberly (1994); Bertola et al. (2005). In the standard textbook (S,s) model, inventories are allowed to vary between two target levels of stock, the upper, S, and the lower, s. Applied to consumer spending, the upper band S and the lower band s refer to the maximum and minimum stock of consumption goods a household desires to hold relative to their income. The consumer increases spending when stocks fall to the lower target level and reduces it when they reach the upper level. In this framework, the existence of transaction costs associated with purchases of durables leads to an irregular adjustment particularly for this category of goods. Lam (1991) considers a situation when the consumer uses a threshold criterion instead of adjusting the stock regularly in small increments to close the gap between desired and actual level of durables. In his model the consumers buy or sell durables when the stock exceeds a certain lower or upper threshold, but do not act otherwise. He shows that households adjust stocks infrequently and expenditures react more to large income shocks than to small shocks. Furthermore, desired stocks are not very sensitive to transitory income, upward adjustments happen quicker than downward adjustments and thresholds levels vary across households. He attributes the different speed of adjustment to imperfections in resale markets while the threshold heterogeneity across households is a consequence of liquidity constraints. When households are faced with a lack of access to credit or excessively expensive �nance, the upper threshold level increases as they are unable to adjust their durables stock to the desired higher level. The low threshold on the other hand will be small because if the desired stock is lower than the actual stock, households might be tempted to resell their durables in order to be able to purchase the nondurables they cannot access due to credit constraints. Attanasio (2000) provides additional insights into the relationship of the adjustment bands and consumer income. He models the purchases of cars by US households as an (S,s) rule de�ning his thresholds in terms of the ratio of durables to non-durables. Using a large microeconomic dataset he directly estimates the (S,s) parameters and �nds that desired ratio of durables is lower for households with lower income. In addition, he also �nd evidence that the (S,s) bands are large. This implies pervasive inertial behavior and potentially high relevance for the determination of aggregate expenditure. Unfortunately, as shown by the paper itself, explicitly deriving aggregate implications proves difficult. Such aggregate implications are attempted in Eberly (1994). Similarly to Attanasio (2000), this paper 6 tests for (S,s) bands behavior under uncertainty using data at the level of the individual household. Based on Grossman and Laroque (1990), her (S,s) model assumes that households adjust their stock of durables to a desired share of wealth. Once this level of durables to wealth stock is achieved, households allow it to depreciate until it reaches a critical lower bound level. At this point they purchase a new durable. She distinguishes between liquidity constrained and non-liquidity constrained households and focuses only on those households facing an adjustment cost. She �nds strong evidence for the (S,s) type adjustment which she also expects to translate to the aggregate level. However, according to Foote et al. (2000), the (S,s) type adjustment model has very little to tell at the aggregate level, because any discrete adjustments will be smoothed out due to the presence of agent heterogeneity. Like previous studies, Bertola et al. (2005) also attempt to derive aggregate implications from effects observed at the micro-level and indeed �nd that the negative effects of uncertainty on individual house- holds translates into a negative aggregate effect. Speci�cally, they assess the effect of uncertainty on the frequency and size of adjustments for three different types of durable goods: vehicles, furniture, and jewelry. They show that small adjustment costs can imply wide ranges of inaction under uncertainty. Higher uncertainty widens the range of inaction and a more uncertain future leads to a lower probability of adjustment, but the adjustment is larger if it does occur. They �nd evidence that lump-sum adjust- ment costs are the predominant source of inaction in their dataset. Most of their results only hold for the vehicles, most likely because the other two categories are less subject to adjustment costs. Other studies have instead focused on the impact of uncertainty on consumer behavior speci�cally during downturn periods. In a study closely related to this one, Romer (1990) looks for evidence of the uncertainty hypothesis in the 1930’s recession, using data for the United States. The paper assumes the existence of an inverse relationship between consumer spending on durable goods and uncertainty about future income. The paper also suggests that one should expect a positive wealth effect on non-durable goods: consumers who are not buying durable goods will have more wealth to spend on perishable goods. She establishes that uncertainty is a positive function of stock market volatility. On this basis, stock market volatility and consumer spending on durables should be negatively related. The choice of a stock market volatility measure as a proxy of uncertainty opens, however, the possibility for alterna- tive explanations. Therefore, she also explores alternative sources of the nexus between stock market volatility and spending on durables, namely 1st order stock market adjustment and �nancial constraints. However, even controlling for those effects, the uncertainty hypothesis remains valid. Carroll and Dunn (1997) analyze the recession of 1990s that was characterized by a spontaneous decline in consumption, particularly in durables. The recession came after a period of strong build-up of household debt, in part driven by high spending on durable goods, which leads them to explore the link between uncertainty and balance sheet deterioration. They �nd that unemployment expectations influence spending beyond any information those expectations contain about future levels of income. When uncertainty increases consumers postpone durable purchases until balance sheet conditions improve. 7 2.3 Uncertainty and aggregate trade Previous literature does not suggest a consensus about how the (S,s) type micro level behavior will translate to the aggregate level. As outlined in the previous section, while Eberly (1994) and Bertola et al. (2005) claim that there are strong aggregate implications resulting from uncertainty faced by individual consumers, Foote et al. (2000) are rather sceptical about a sizeable aggregate adjustment. However, the latter study does not consider a situation in which the whole economy faces a major shock and all agents are suddenly subject to high levels of uncertainty. Under these circumstances one could reasonably expect that a high aggregate effect will be observed if all agents that were due to adjust in the period of the shock suddenly decide to postpone their invest- ments and consumption decisions. The real-option effect from increased uncertainty over economic and business conditions thus is likely to cause an initial and sudden drop in activity as many agents respond to the mechanisms described previously and pause investment, hiring and consumption at once. As the uncertainty subsides there might be an overshoot in activity arising via both, the producer and consumer uncertainty channel. Connecting this to trade, it is uncertainty on the buyer side that matters. In the presence of domestic uncertainty we should therefore not expect a large impact on exporters unless the uncertainty abroad is correlated with domestic uncertainty. If the importers are not subject to shocks themselves, the most affected producers will be those focusing on domestic demand. Theoretically, if the period of heightened domestic uncertainty is protracted, local producers could potentially start redirect- ing their sales toward foreign markets and thus boost exports. On the other hand, the consequences of high uncertainty abroad can be very damaging for exporters. A big drop in durable consumption and investment goods purchases may have very strong aggregate consequences for exporters specializing in this type of goods. Based on the insights from the literature on the impact of uncertainty on investment and consumer spending, we can reasonably expect the impact of uncertainty on trade to be non-linear. This non- linearity is likely to arise for two reasons. The (S,s) model suggests that trade will react only if the uncertainty shock is sufficiently high. Second, because of consumer heterogeneity uncertainty needs to affect most producers and consumers to trigger a trade reaction. As pointed out by Foote et al. (2000) not all consumers adjust at the same time. Therefore, in relatively normal times when only some of the consumers face uncertainty about their future income stream and employment prospects and the aggregate uncertainty is relatively low, the aggregate adjustments should happen relatively regularly and smoothly. However, at times of extreme uncertainty the adjustment bands for most consumers will widen and we are more likely to observe a substantial overreaction of durable purchases while non- durables should be affected only marginally. Major uncertainty shocks often overlap with other turbulent events characterized by stock market crashes reducing the wealth of the economy and reducing access to �nance. Previous studies imply that both should have implications for the behavior of economic agents when making decisions about adjusting their stocks of durables or investment goods. As these purchases are costly to reverse, consumers and producers are likely to wait when their access to �nance gets restricted. Similarly, a reduction in wealth will imply that agents adjusting their stocks in proportion to wealth will now have a higher than desired 8 stock and will thus prefer to wait or even attempt to disinvest from their existing stock. Therefore, exporters, and particularly those focusing on sales of goods that are hard to resell, should expect a further reduction in trade if the importer uncertainty is accompanied with a major hit to the real economy. To summarize, based on guidance from previous literature we expect the following parameters to mat- ter for the aggregate response of trade in the downturn and in the recovery. First, importer uncertainty should matter more for exporters than domestic uncertainty. Second, the size of the shock should also matter, with a potentially non-linear effect. Third, liquidity constraints and large adjustments in wealth, that tend to accompany major uncertainty shocks, will be aggravating the contraction of expenditures. Finally, the adjustment costs should also matter. We expect that countries specialized in goods entailing high adjustment costs (such as durables or investment goods) should experience an heightened impact of uncertainty on exports. Exports of non-durables on the other hand could potentially be boosted due to the fact that consumers have a higher disposable income that they can devote to these goods, due to the savings on durable goods. 3 Producer and consumer uncertainty: VAR analysis As a preliminary test of the hypotheses outlined above we focus on US data and conduct a vector autoregression (VAR) analysis analogous to that of Bloom (2009). He estimates a standard VAR model that includes stock market volatility used as a proxy for uncertainty and the following additional variables: the S&P500 stock market index, federal funds rate, average hourly earnings, consumer price index, hours in manufacturing, employment in manufacturing and industrial production. His sample covers the period June 1962 to June 2008 which comprises seventeen uncertainty shocks that are depicted in Figure 1. For reasons discussed in Section 4.2 of this paper he works with a stock market volatility based measure of uncertainty. Rather than working with stock market volatility explicitly he chooses to only work with periods of exceptional uncertainty during which the measured volatility exceeded the mean by at least 1.65 times the standard deviation. The identi�cation of the uncertainty shock is achieved through a standard Cholesky decomposition with the uncertainty shocks ordered after the stock market index so that the effect of the stock market is already accounted for when looking at the effect of the uncertainty shocks. In order to extend Bloom’s analysis, which only focuses on the impact of uncertainty of production, we augment this baseline VAR in several ways. To assess the impact of uncertainty shocks on consumption we add consumption expenditures in the model. To further disentangle the effects of consumer uncertainty, we split the consumption expenditures into a durable and non-durable component. Finally, as the focus of this paper is on international trade, we augment the VAR with monthly import and export levels. Given that many uncertainty periods occur at times of major �nancial constraints and stock market crashes we also take a brief look at the impact of these effects on our variables of interest. We use the same dataset as in Bloom (2009). Additional variables, not included in Bloom’s dataset, are taken from the Federal Reserve Economic Data (FRED) of the St. Louis Fed. 9 Figure 2 reproduces Bloom’s original results. Unlike in his paper we use 4 lags instead of 12. We make this choice based on the information criteria that suggest using between three and �ve lags with the Schwarz Bayesian criterion pointing to four. However, this decision does not substantially affect the results of the VAR reported by Bloom. Figure 2 shows that in response to a one standard deviation increase in uncertainty industrial output initially drops down. As Bloom further explains with his theoretical model this is because the inaction bands of �rms increase leading to a freeze in activity. As uncertainty retrenches a burst of activity takes place. Because of the freeze and burst in activity, recovery takes the form of a rapid catching up phase and a temporary overshoot over the medium term. To make up for the shortfall in investment, exacerbated by the depreciation of existing goods and capital, �rms spend at a faster pace than usual. Hence, there is an initial overshoot in production activity. However, over the long term, industrial production returns to trend. The responses to �rst order stock market shocks and to shocks to the federal funds rate are also in line with Bloom’s original results. Stock market increases affect production positively and hikes in federal funds rate lead to protracted drops in industrial activity. To shed light on the consumer behavior under uncertainty we augment the baseline VAR with con- sumer expenditures. Figure 3 shows the impulse responses of industrial production and consumption expenditure to one standard deviation increases in uncertainty, stock market index and federal funds rate. The reactions of the industrial production are consistent with the results in Figure 2. Interestingly, when comparing the reactions to those of industrial production the aggregate consumption expenditures follow a very similar pattern in response to all three shocks. Consumption drops in response to uncertainty and rises over trend during the recovery, however, the overshoot is less pronounced. Unsurprisingly a rise in stock market that makes consumers wealthier increases aggregate expenditures. In contrast, a monetary policy contraction that leaves consumers �nancially constrained leads to a protracted drop in spending. Most of the existing literature on the real effects of uncertainty suggests that durables are likely to be the segment of consumption most elastic to uncertainty shocks. We test this prediction by estimating a second VAR speci�cation, where consumption expenditure is accounted for by distinguishing between the durable and a non-durable components. Figure 4 shows the impulse response functions derived from this estimation. The results suggest that the initial drop in consumption expenditures observed in response to an uncertainty shock is mainly driven by durable expenditures that are also the main driver of the medium term overshoot. However, it also seems that the protracted nature of the drop is driven mostly by non-durables, which do not drop as much as durables, but stay below trend longer. Both, durable and non-durable consumption, responds positively to favorable stock market shocks which implies that consumers do not spend disproportionately more on one of the categories in response to an increase in wealth. However, the picture changes when looking at the reactions to the federal funds rate. A monetary contraction leads to a protracted drop in durable consumption suggesting that American consumers secure most of it using loans. In contrast, the same rate hike boosts the consumption of non- durables. The increase in non-durable consumption could be explained by consumers substituting away from durable consumption that becomes too expensive due to the rising cost of �nance. This implies that in crisis periods, such as the one following the Lehman bankruptcy, during which excessive levels of 10 uncertainty are combined with a major credit crunch one can expect a high initial discrepancy between durable and non-durable consumption due to both �nancial constraints and elevated uncertainty levels. The �nal set of results shown in Figure 5 documents the response of international trade to uncertainty. We augment Bloom’s original speci�cation with the effective exchange rate and aggregate imports and exports. We do not include consumption expenditures in this estimation. Because of monthly data availability the sample is almost 20 years shorter, starting only in January 1980. Reassuringly, despite using a shorter sample and omitting some of the shocks, the response of industrial production is almost unchanged. As shown in Figure 5 the reactions of aggregate imports and exports to a rise in domestic uncertainty differ considerably. Exports experience a short sharp increase after which they drop almost immediately to their previous path. In contrast, imports drop dramatically and the negative response is protracted. We do not have sufficiently long monthly time series to estimate the impact on the durables and non-durables components of trade separately.5 Therefore, based on previous �ndings, one could speculate that most of the initial drop is driven by the durable component of imports while the protracted nature is primarily due to the non-durable part. Both imports and exports respond very similarly to changes in the stock market index. The increases in wealth leads to a rise of both above trend for an extended period. A contractionary monetary policy leads to an initial short lived rise in aggregate imports that soon converts into a protracted drop. A potential explanation for this is that initially the appreciation brought by a rise in interest rates makes foreign goods cheaper, however, over time the restrictive effects induced by �nancial constraints take over and reduce imports. Surprisingly, exports also rise in response to a hike in the federal funds rate. Even though, one would expect a negative effect on exporters via �nancial constraints and appreciation, a potential explanation for this result is that the biggest exporters responsible for most of the aggregate exports are less �nancially constrained and with the contraction of the domestic demand might be more aggressive in selling their products abroad. In sum, it seems that uncertainty matters in an international setting, particularly for imports. In the subsequent sections we look at whether the preliminary results provided by the VAR analysis conducted on US data carry through to an international setting. Speci�cally, we will be focusing on the impact on exports of importer uncertainty. This, based on the preliminary results, is expected to be more important for exporters than uncertainty experienced at home. 4 Importer uncertainty in bilateral trade data 4.1 Methodology In accordance with the theoretical insights summarized above and with results from the VAR analysis, we focus on the effects of importer uncertainty on trade flows over the period 1990-2009. We examine the evolution of quarterly trade flows by means of a model of bilateral trade. The use of bilateral trade allows us to assess the effect of foreign uncertainty while controlling for domestic developments and additional 5 To our knowledge, the only cross-country dataset reporting monthly (and quarterly) bilateral trade data is Trade Map of the International Trade Centre (ITC). Market Access Map, Market Analysis Tools, International Trade Centre, www.intracen.org/marketanalysis which provides data series starting in 2005 only. 11 factors that might have affected the conditions of the importing economy during uncertainty periods. The bilateral speci�cation also helps reduce endogeneity concerns as it is unlikely that trade with one country would affect aggregate importer uncertainty. We choose a dynamic speci�cation because there are strong economic reasons to believe that there is persistence in trade. Bun and Klaassen (2002) list established distribution and service networks and habit formation of consumers as two major reasons why yesterday’s trade is a good predictor of trade with the same trading partner today. Guiso et al. (2009) add trust as an additional reason, �nding that establishing trust with a foreign country leads to an increase of trade by 10%. Finally, two older papers by Dixit (1989b) and Baldwin (1988) �nd that there is hysteresis in trade due to sunk costs. Due to all these reasons estimation of a static model would lead to an autocorrelated error and incorrect inference. Therefore, we instead estimate the following baseline autoregressive distributed lag (ARDL) speci�cation: n1 n2 n3 n4 n5 Xodt = αj Xodt−j + βj Udt−j + γj Ydt−j + δj RERodt−j + φj REERodt−j + θot + ηod + εodt j =1 j =0 j =0 j =0 j =0 where Xodt are exports from country of origin o imported by destination country d in time t, REER is the real effective exchange rate, RER is the bilateral exchange rate, Y is demand and U refers to uncertainty. In addition, we use two sets of �xed effects. The set of exporter time dummies θot controls for developments in the exporter country such as current demand levels or �nancial conditions. Country pair effects ηod are included to account for any pair speci�c characteristics that do not vary over time - for example distance or common language would fall in this category. We treat the effects as �xed rather than random in order to allow for correlation with other regressors. To analyze the effects of �rst order adjustments and �nancial constraints that might accompany uncertainty periods, in further analyzes we augment our baseline speci�cation with proxies for these two effects. Given the relatively long time dimension we are working with, an OLS estimation is unbiased and consistent as long as εodt is white noise. We choose the ARDL estimation method rather than the commonly used GMM because of its preferable properties for our panel structure. While GMM has been developed for small T panels and requires stationary data, our included regressors are highly likely to be very persistent. In order for a valid estimation of the above speci�cation we will need to make sure that any autocorrelation is removed from the error. Keeping this concern in mind we determine the optimal number of lags by testing down from a general speci�cation. Our focus throughout the estimation is on the impact of uncertainty on the long run levels of trade. This means that rather than looking at the impact of the individual contemporaneous or lagged terms we will instead focus on the cumulative effect uncertainty shocks (and the remaining regressors) have on the long run level of exports. The uncertainty multiplier, or what we shall refer to as the long run effect n2 βj in the following sections, is computed as 1− n1 and analogously for the remaining regressors. j =1 αj 12 4.2 Data 4.2.1 Bilateral trade data The dependent variable captures the aggregate exports between country of origin o and destination country d. The trade data to construct the dependent variable are taken from the IMF Directions of Trade Statistics. They cover the period from the �rst quarter of 1990 to the last quarter of 2009. We work with a set of 32 developed and developing countries. The sample selection is largely dictated by data availability, but we also aim at having a representative sample. 4.2.2 Uncertainty Similarly to Bloom (2009) we use a measure of stock market volatility as a our main proxy for uncertainty. Implied volatility such as VIX would be the preferred measure, because it better reflects the sentiments about near future. Such measure is not readily available for a large set of countries and therefore we will be using the actual volatility of the stock market in the importing country. However, Bloom (2009) demonstrated in the example of the US, that for the subsample in which both the actual and implied volatility are available the correlation exceeds 0.8. This very high correlation should make the actual volatility an acceptable proxy. An additional argument in favor of using stock market volatility is also that this measure is correlated with many alternative measures of uncertainty used in the empirical literature. Alexopoulos and Cohen (2009) show that the variance of the stock market is highly correlated with more pragmatic measures, such as the number of times a major newspaper mentions the word uncertainty within a given period. Moreover, Bloom (2009) reports statistical evidence showing that stock market volatility is strongly linked to other measures of productivity and demand uncertainty, including the variance of �rm pro�t growth and TFP growth and the disagreement among professional forecasters in expectations about macroeconomic variables. In our estimations, instead of using the stock market volatility explicitly, we employ dummy variables identifying periods of exceptional uncertainty. We adopt the method suggested by (Bloom, 2009), i.e. our dummy variable takes a value of one when the Hodrick-Prescott detrended stock market volatility exceeds its mean by at least 1.65 standard deviations.6 Table 1 documents the episodes of exceptional uncertainty included in our sample of countries. It shows that during the last two decades there have been numerous uncertainty shocks in developing and advanced countries alike. Most of them coincide with times of �nancial turmoil, most notably during the crises in Asia, Russia and Mexico or with political shocks, as in 2001. 4.2.3 Demand We use the total imports of the destination country d minus the imports from the exporter o as an approximation for the absorption capacity of the importer. Subtracting the imports from the exporter avoids endogeneity problems, as suggested by (Bricongne et al., 2012). We prefer this measure to GDP 6 This choice might seem somewhat arbitrary and we will be looking at different thresholds in our estimations. 13 partially because of data availability and partially because it is a more representative measure of the demand for foreign goods. 4.2.4 Exchange rates Our speci�cation includes two different controls for exchange rates: the real bilateral exchange rates RER and the real effective exchange rate REER. The RER is computed as a product of the bilateral nominal exchange rate and the ratio of importer to exporter CPI. Both, CPI indices and bilateral exchange rates, are taken from the IMF International Financial Statistics (IFS). We derive the nominal rates using the exchange rates against the US dollar. The choice of the CPI over the PPI or other producer related indices is due to the wider data availability. The RER is de�ned so that an increase means a depreciation, i.e., a boost in competitiveness. Hence, we can expect the overall effect of this variable on trade will be positive. The REER measure captures the real effective exchange rate of importer d in relation to all the trade partners except the exporter o. The effect of this variable is a priori ambiguous. An increase in this measure indicates that the real exchange rate of the importer appreciates on average towards all trade partners and competitors of exporter o. This implies that their exports to market d are cheaper and one could reasonably expect that if the exports from o are substitutable with exports from other countries, the importer will switch to new suppliers. In this case the overall effect would be negative, to signal the substitution effect. However, the sign could also be positive and signal a wealth effect. The more favorable exchange rate could mean that importer d disposes of more income to buy goods from all trade partners, including exporter o. Hence, the sign of REER remains largely an empirical question. A positive sign indicates that the wealth effect prevails over the substitution effect while a negative sign that the substitution effect dominates. 4.2.5 Financial controls Despite the fact that it is common practice to quantify uncertainty using measures of �nancial market volatility, there are some obvious concerns. Given the �nancial nature of the proxy for uncertainty, it is not possible to completely rule out an alternative story. It might well be that what depresses economic activity and demand is not uncertainty per se but a rise in �nancial frictions or a reduction in wealth. The uncertainty periods often occur during times of �nancial crises that tend to be characterized by �nancial frictions that make the �nancing of export related costs harder. To account for this possibility we augment our baseline speci�cation by a proxy for �nancial constraints. The spread between the money market rate and treasury bill rate (ted) is a commonly used measure of credit crunch, however, because of data availability we use the level of the money market rate. The two measures are very highly correlated for the subsample for which we have both ted and the money market rate, allowing us to construct the spread. In addition to �nancial constraints, we also control for wealth effects, which we measure as the change in the level of real stock prices. The rationale for this is that movements in stock prices can have at the same time a wealth effect as well as an uncertainty effect. In periods of lack of con�dence, a drop in 14 consumer and investment spending and therefore trade may also be due to the decline in stock prices. In other words it is possible that a con�dence crisis depresses consumption, investment and trade simply by destroying a great deal of wealth. 5 Results 5.1 Baseline results The estimation results of our baseline speci�cation are presented in Table 2. We have determined the optimal lag length by using the Schwarz Bayesian and Akaike criteria. They both suggest that the best speci�cation is one that includes four lags of the dependent variable and the demand measure and two lags of the remaining variables. The criteria are very similar for a speci�cation that includes four lags of all variables. Therefore, in the baseline results we show them both for comparison. Results of the Lagrange Multiplier test implies that this lag length also removes autocorrelation from the residuals. Column (1) shows results for a speci�cation with two lags on both exchange rate measures and uncertainty and column (2) reports a speci�cation with four lags. None of the third or fourth lags are signi�cant, therefore, we will stick to using two lags throughout the rest of the paper. As shown in Table 2 the sum of the coefficients on lags of the dependent variable is lower than one which implies that the dynamic relationship is stable. The effect of uncertainty is consistently negative in all lags, however, the strongest and most signi�cant effect appears to be contributed by the �rst lag, i.e. with a one quarter delay. The long run effects of all variables are summarized in Table 3. Panel I shows that when including two lags our main variable of interest, the uncertainty shock faced by an importer, leads to a highly signi�cant 11.5% reduction of aggregate exports. The negative effect increases further to 15% when including four lags of uncertainty. The effect of the remaining control variables on trade is in line with what one could expect. An increase in demand leads to an almost one for one increase in exports. A depreciation of the bilateral exchange rates boosts trade, however, the effect is statistically insigni�cant. Finally, a depreciation of the real effective exchange rate leads to a drop in exports suggesting that the substitution effect described above dominates over the income effect. Because the 2008/2009 shock has been the largest uncertainty shock experienced by most countries in our sample we also check to what extent this single period drives our results. Panels III and IV of in Table 2 and of Table 3 report the results obtained when limiting the sample to the pre-2007 period only. This restriction leaves the results almost exactly the same suggesting that the extreme drop in international trade observed after the Lehman Brothers bankruptcy and the subsequent return to a relatively strong path of recovery was proportional to the exceptional size of the uncertainty shock that triggered it. Hence, while quantitatively speaking the effects were much stronger, qualitatively speaking, the relationship between uncertainty and trade during the most recent crisis did not post structural differences compared to past episodes. 15 5.2 Financial constraints and wealth effects Uncertainty periods often overlap with major stock market crashes and �nancial crises. In order to see if our uncertainty measure is just capturing the reduction in wealth or �nancial constraints, which would both lead to a reduction in imports, we augment our baseline speci�cation with proxies for these two effects. The results of the estimation including the money market rate and a the adjustment in the level of real stock prices, that we refer to as wealth, are reported in Table 4. We �rst include each measure separately and then we estimate a speci�cation that includes both of them. The speci�cations in columns (1) to (3) show the results using the whole sample while the last three columns refer to a subsample with the last quarter of 2006 as cut-off. Exchange rates and demand are included in all estimations. The results are very similar to the baseline case, however we do not report them for the economy of space. The long run effects of all variables are summarized in Table 5. They show that including wealth has very little bearing on the estimated effect of uncertainty. The long run effect of the uncertainty shock stays almost exactly the same while the wealth effect itself turns out only marginally signi�cant when included without a measure of �nance and highly insigni�cant otherwise. The situation changes when looking at the impact of �nancial constraints. The effect of a rise of the money market rate, while highly signi�cant, has only a very small economic effect. However, its inclusion reduces the impact of an uncertainty shock by about one third - to a 7.6% drop in exports. 5.3 Non-linearities In the estimations so far we have been following Bloom(2009) in de�ning the uncertainty shock as mean stock market volatility plus 1.65 times the standard deviation as a cut-off. The purpose of this section is twofold. First, it assesses whether the results are robust to other de�nitions of the cut-off. Second, using different cut-off levels provides insights into whether or not - as the previous literature is suggesting - there is non-linearity in the response of trade to uncertainty. We work with four different cut-off levels starting with 0.5 and increasing in increments of 0.5 up to two times standard deviation. The periods when uncertainty increases half a standard deviation above its mean are relatively frequent while there are no periods in which uncertainty exceeds the mean by more than two and a half standard deviations. The results of the baseline speci�cation and a speci�cation including the �nancial constraints and wealth effects are shown in Table 6. The corresponding long run effects are summarized in Table 7. The impact of uncertainty shocks on trade is consistently negative regardless of the chosen cut-off. However, for the lowest two cut-off levels it becomes statistically insigni�cant when the �nancial constraints and wealth controls are also included. Strikingly, the impact on trade is much more adverse when moving from the 1.5 cut-off to the highest considered cut-off level than for any other increment of the same size. Overall the results suggest that in order to see a large adverse effect on trade the uncertainty has to reach very high levels implying that the effect on trade is indeed non-linear. 16 5.4 Durables vs. non-durables Much of the previous literature implies that the effect of an uncertainty shock should be channeled primarily through durable consumption, although there is no consensus on the extent to which the relationship at the microeconomic level translates into macroeconomics. In order to assess this nexus we would need bilateral monthly data by product category. As discussed earlier in the paper, these data are not available for a wide set of countries and for a sufficiently long time span. Hence, we use an indirect method to categorize trade by its durability. We compute the share of durables exported by exporter o to importer d as a share of total imports of d . We identify the shares of durables by combining ISIC3 industry level trade data from Comtrade with the classi�cation of durable and non-durable industries compiled by Kroszner et al. (2007). Then we order all the relationships of an importer by their durability, split them into quartiles and estimate separate regressions for each quartile. In our de�nition the top quartile refer to relationships most intensive in durables trade. For example looking at the US as an importer the relationship most intensive in durables is with Japan from where more than 80% of total imports comprise durable goods. The relationship with lowest durability is with New Zealand that only provides 28% durables in its total exports to the United States. Tables 8 and 9 show that most of the negative effect of uncertainty on trade is generated by relation- ships that fall into the second and third quartile of durability. Uncertainty shocks are found to have an insigni�cant effect in the relationships in the top and bottom quartiles of durability. Without an explicit test of durability that uses industry level data, we cannot unambiguously determine the differential im- pact of uncertainty on goods of different durability. Despite this however, the lack of signi�cance for the effect of uncertainty on the bottom quartile of durability con�rms that the latter is an important dimension for explaining the relationship between uncertainty and trade. On the other hand, the results for the top quartile of durability are more puzzling. There are several possible explanations for the lack of signi�cance of this set of trade flows in our data. Supply chain considerations, the possibility that relationships with highest durability are also important in absolute terms or compositional and substitu- tion effects may have a bearing on the results. For example importers facing an uncertainty shock on the domestic market may be securing the most durable goods because these are fundamental components to their exports in the context of supply chains. Alternatively, consumers may be substituting from more durable to less-durable goods bought from the same partner with the result that the aggregate bilateral trade between two countries remains unaffected. In combination with the VAR analysis, these results suggest that the relationship between uncertainty shocks and durability deserves further analysis concerning the size and timing of both the contraction and the recovery pattern. 5.5 Prior experience with uncertainty shocks To conclude our analysis we assess whether prior experience or more generally high frequency of shocks affects the reaction of importers to uncertainty. One could expect that countries that are often subject to volatility would react in a less extreme way to yet another uncertainty shock.7 However, as Tables 10 7 The limitation of this analysis is that prior experience with uncertainty shocks might already be reflected in the level of volatility itself, i.e. in a country with frequent shocks the same event might lead to a lower increase in volatility compared 17 and 11 show, having shocks more often has a more pronounced negative effect on trade as countries with a relatively higher incidence of shocks over the sample suffer a higher loss in trade. When, instead of splitting the sample, we explicitly control for the number of shocks the importer experienced in the past the results show almost no reduction in the long run coefficient of uncertainty. In sum, the incidence of uncertainty shocks or extensive prior experience with uncertainty does not seem to mitigate the effects it has on trade. 6 Conclusion The large drop in international trade observed at the end of 2008 and the beginning of 2009 has gener- ated considerable attention worldwide. Many commentators invoked uncertainty and the ’wait and see’ attitude that followed as major factors in these turbulent events. The uncertainty hypothesis predicts that when uncertainty is sufficiently high economic agents will postpone purchases, in particular of such goods that are impossible or very costly to resell. In this paper, we explore the hypothesis by focusing not only on the most recent major uncertainty event but also similar past events in a set of 32 developed and developing countries. We ask if, when controlling for other factors that tend to accompany major uncertainty shocks, uncertainty has power for explaining drops in trade. In line with the previous lit- erature we use stock market volatility as a proxy for uncertainty and focus exclusively on periods when the measured volatility was exceptionally high. Based on theory predictions and our preliminary VAR results, we expect domestic uncertainty to affect mostly producers focused on domestic demand, while domestic exporters will be mainly affected by uncertainty abroad. To quantify the impact of importer uncertainty on their foreign suppliers, we estimate a bilateral dynamic model of trade in which we control for the developments in the exporting economy by means of �xed effects. Our main results are the following. Uncertainty in the importer country has a strong negative effect on countries’ exports. The impact gets smaller, but does not disappear, when controlling for aggravating factors such as �nancial constraints and wealth adjustments. We further �nd that uncertainty shocks affect trade flows in a non-linear fashion, i.e., they need to reach a certain threshold level in order to translate into strong aggregate effects on trade. Uncertainty becomes particularly relevant, when its levels are unusually high, such as in the 2008-2009 crisis of con�dence. Con�rming what previous literature predicts, our results suggest that adjustment costs matter. Countries specialised in non-durable and investment goods, which entail little adjustment costs, are not affected by uncertainty. However, a more thorough industry level analysis is needed to further prove the role of durability, as our results indicate that the negative effect of uncertainty on trade stems not from the trade relationships most intensive in durable goods, but from those with a more balanced mix between durable and non-durable goods. Interestingly, our results show that the effect of uncertainty in the post Lehman crisis did not post new structural features. The strong trade reaction was due to the size of the uncertainty shock but not to a change in the mechanisms regulating the relationship. Indeed, when excluding the most recent period from the sample we �nd almost exactly the same elasticity coefficients. Furthermore, we do to a country whose consumers do not have any major uncertainty event in their recent memory. 18 not �nd evidence of learning from past shocks, because even importers with substantial past experience with uncertainty shocks overreact when faced with a new shock. A caveat here is that, because we are measuring uncertainty in terms of stock market volatility, the experience with past shocks might already be reflected in the volatility increase itself. This would mean smaller increase in stock market turmoil in a country with high experience when compared to a country that has not experienced many extreme uncertainty events in the past, when faced with the same shock. Still, our evidence suggests that should we face a new uncertainty shock of a similar magnitude, the aggregate effects on the real economy could again be substantial. Therefore, striving for a stable economic environment free of major volatility spikes, should stay very high on the agenda of policy makers worldwide. 19 References Alexopoulos, M. and J. Cohen (2009, February). Uncertain times, uncertain measures. Working Papers tecipa-352, University of Toronto, Department of Economics. Arrow, E., T. Harris, and K. Marschak (1951). Optimal inventory policy. Econometrica (19), 250–272. Attanasio, O. P. (2000, October). Consumer durables and inertial behaviour: Estimation and aggregation of (s, s) rules for automobile purchases. Review of Economic Studies 67 (4), 667–96. Baldwin, R. (1988, September). Hyteresis in import prices: The beachhead effect. American Economic Review 78 (4), 773–85. Benassy-Quere, A., Y. Decreux, L. Fontagne, and D. Khoudour-Casteras (2009). Economic crisis and global supply chains. Working Paper 15, CEPII. Bernanke, B. (1985, January). Adjustment costs, durables, and aggregate consumption. Journal of Monetary Economics 15 (1), 41–68. Bernanke, B. S. (1983, June). Nonmonetary effects of the �nancial crisis in propagation of the great depression. American Economic Review 73 (3), 257–76. Bernanke, B. S. (1984, August). Permanent income, liquidity, and expenditure on automobiles: Evidence from panel data. The Quarterly Journal of Economics 99 (3), 587–614. Bertola, G., L. Guiso, and L. Pistaferri (2005, October). Uncertainty and consumer durables adjustment. Review of Economic Studies 72 (4), 973–1007. Blanchard, O. (2009). (Nearly) nothing to fear but fear itself. The Economist, 29 January. Bloom, N. (2009, 05). The impact of uncertainty shocks. Econometrica 77 (3), 623–685. Bresnahan, T. F. and D. M. G. Raff (1991, June). Intra-industry heterogeneity and the great depression: The american motor vehicles industry, 1929-1935. The Journal of Economic History 51 (02), 317–331. Bricongne, J.-C., L. Fontagne, G. Gaulier, D. Taglioni, and V. Vicard (2012, May). Firms and the global crisis: French exports in the turmoil. Journal of International Economics 87 (1), 134–146. Bun, M. and F. Klaassen (2002). The importance of dynamics in panel gravity models of trade. Discussion Paper 2002/18, Universiteit van Amsterdam. Bussiere, M., G. Callegari, F. Ghironi, G. Sestieri, and N. Yamano (2011, December). Estimating trade elasticities: Demand composition and the trade collapse of 2008-09. NBER Working Papers 17712, National Bureau of Economic Research, Inc. Caballero, R. (2009a). A global perspective on the great �nancial insurance run: Causes, consequences, and solutions (part 2). VoxEU.org, 23 January. Caballero, R. (2009b). Sudden �nancial arrest. VoxEU.org, 17 November. 20 Carroll, C. and W. Dunn (1997, May). Unemployment expectations, jumping (s,s) triggers, and household balance sheets. In NBER Macroeconomics Annual 1997, Volume 12, NBER Chapters, pp. 165–230. National Bureau of Economic Research, Inc. Cheung, C. and S. Guichard (2009, October). Understanding the world trade collapse. OECD Economics Department Working Papers 729, OECD, Economics Department. Dixit, A. (1989a). Entry and exit decisions under uncertainty. The Journal of Political Economy 97 (3), pp. 620–638. Dixit, A. (1989b). Hysteresis, import penetration, and exchange rate pass-through. The Quarterly Journal of Economics 104 (2), pp. 205–228. Dixit, A. K. and R. S. Pindyck (1994). Investment under uncertainty / Avinash K. Dixit and Robert S. Pindyck. Princeton University Press, Princeton, N.J. :. Eberly, J. C. (1994, June). Adjustment of consumers’ durables stocks: Evidence from automobile pur- chases. Journal of Political Economy 102 (3), 403–36. Foote, C., E. Hurst, and J. Leahy (2000, May). Testing the (s, s) model. American Economic Re- view 90 (2), 116–119. Grossman, S. J. and G. Laroque (1990, January). Asset pricing and optimal portfolio choice in the presence of illiquid durable consumption goods. Econometrica 58 (1), 25–51. Guiso, L., P. Sapienza, and L. Zingales (2009, August). Cultural biases in economic exchange? The Quarterly Journal of Economics 124 (3), 1095–1131. IMF (2004). Exchange rate volatility and trade flows - some new evidence. Technical report, International Monetary Fund. Kamil, H. and K. Rai (2009). On the eve of retrenchment? The effect of the global credit crunch on foreign banks’ lending to emerging markets. mimeo. Kroszner, R. S., L. Laeven, and D. Klingebiel (2007, April). Banking crises, �nancial dependence, and growth. Journal of Financial Economics 84 (1), 187–228. Lam, P.-S. (1991, February). Permanent income, liquidity, and adjustments of automobile stocks: Evi- dence from panel data. The Quarterly Journal of Economics 106 (1), 203–30. Levchenko, A. A., L. T. Lewis, and L. L. Tesar (2010, May). The collapse of international trade during the 2008-2009 crisis: In search of the smoking gun. NBER Working Papers 16006, National Bureau of Economic Research, Inc. Ohanian, L. E. (2002). Why did productivity fall so much during the great depression? Quarterly Review (Spr). 21 Rodrik, D. (1997). Has Globalization Gone Too Far? Number 57 in Peterson Institute Press: All Books. Peterson Institute for International Economics. Rodrik, D. (2011). The Globalization Paradox. Northon and Company, Limited. Romer, C. D. (1990, August). The great crash and the onset of the great depression. The Quarterly Journal of Economics 105 (3), 597–624. Willenbockel and Robinson (2009). Tba. Technical report, TBA. 22 7 Tables and �gures Figure 1: Periods of exceptional uncertainty Source: Bloom(2009) 23 Figure 2: Domestic uncertainty impact on production Note: Response of industrial production to innovations in uncertainty measured by stock market volatility, stock market levels and federal funds rate based on data used in Bloom (2009). The standard VAR includes the S&P500 stock market index, a stock market volatility, federal funds rate, average hourly earnings, consumer price index, hours in manufacturing, employment in manufacturing and industrial production, in this order. 24 Figure 3: Domestic uncertainty impact on consumption Note: Response of industrial production and consumption expenditure to innovations in uncertainty measured by stock market volatility, stock market levels and federal funds rate based on data used in Bloom (2009). The standard VAR includes the S&P500 stock market index, a stock market volatility, federal funds rate, average hourly earnings, consumer price index, hours in manufacturing, employment in manufacturing and industrial production, and consumption expenditures in this order. 25 Figure 4: Domestic uncertainty impact on durable consumption Note: Response of industrial production and durable and non-durable consumption expenditure to innovations in uncertainty measured by stock market volatility, stock market levels and federal funds rate based on data used in Bloom (2009). The standard VAR includes the S&P500 stock market index, a stock market volatility, federal funds rate, average hourly earnings, consumer price index, hours in manufacturing, employment in manufacturing and industrial production, and durable and non-durable expenditures in this order. 26 Figure 5: Domestic uncertainty impact on trade Note: Response of industrial production and exports and imports to innovations in uncertainty measured by stock market volatility, stock market levels and federal funds rate based on data used in Bloom (2009). The standard VAR includes the S&P500 stock market index, a stock market volatility, federal funds rate, average hourly earnings, consumer price index, hours in manufacturing, employment in manufacturing and industrial production, exchange rate, exports and imports. 27 Table 1: Episodes of exceptional uncertainty in 32 countries, 1990-2009 Year affected countries 1990 13/32 FR,DE,HK,ID(2),IT,JP,MX,NZ,SG,ES,SE,CH,TH 1 (all) 1991 6/32 AR,CN,DE,NZ,PH,ES 2(CN) 1992 10/32 CN,HK,IN(3),JP(2),KR(2),MX,NO,ZA,SE,TH 3(SE) 1993 2/32 CZ,NZ 1 (all) 1994 10/32 CN,CZ,NZ,HK,HU,MY,MX,PL (2),TH,TR(2) 2(PL,TR),3(CN) 1995 4/32 BR(2),CN,MX,PL 2(PL) 1996 2/32 CN,HU 1 (all) 1997 16/32 AU,BR,DK,DE,HK,HU,IT,JP,KR,MY(2),MX,NZ,ZA,ES,TH,TR 2(MY,HU),10(KR) 1998 24/32 AR,BR,CA,DK,FR,DE,HK(3),HU,IT(2),KR(2),MY(2),MX,NO,PH(2), 2(FR,DE,HK,HU,IT,KR,NO,PH,PL,SG,CH,TH,TR), PL,RU,SG(3),ZA(2),ES,SE,CH,TH(3),TR(2),GB 3(PH,RU,ZA,ES) 1999 5/32 BR,HUM,KR,PL,ES 1 (all) 2000 12/32 AU,CA (2),IN,KR,MX,NZ,PH,PL,SNG,ZA,TR (2),US 1 (all) 2001 14/32 FR,DE,HK,IT,JP,KR,NZ,PH,SNG,ES,SE (2),CH,TR,GB 2 for SE 2002 9/32 DK,FR,DE (2),ES,SE (2),CH,TR,GB,US 3(GB),4(FR,CH) 2003 5/32 FR,DE,CH,TR,GB 1(all) 2004 1/32 IN 1 (all) 2005 0/32 – – 2006 5 /32 IN,MX,NO,ZA,TH 1(all) 2007 3/32 AU,PH,SNG 1(all) 2008 28/32 AU(3),BR,CA,CZ,DK,FR,DE,HK(2),HU,IN(2),IT,JP(2),KR,MX, 2(CZ,DK,DE,HK,HU,JP,KR,MX,SG), NO,NZ,PH,PL,RU,SG(2),ZA,ES(2),SE,CH,TH,TR,GB,US 3(AU,BR,FR,IT,RU,ZA,ES,CH,GB) 2009 4/32 IN,IT,GB,US 1(all) Source: Authors calculations. Note: Duration is of one month, unless otherwise stated. Number next to country name indicates no. of episodes in same year. 28 Table 2: Baseline (1) (2) (3) (4) 2 lags 4 lags 2 lags pre 2007 4 lags pre 2007 coef sd coef sd coef sd coef sd L.Exports 0.456*** (0.011) 0.457*** (0.011) 0.451*** (0.012) 0.452*** (0.012) L2.Exports 0.111*** (0.012) 0.109*** (0.012) 0.105*** (0.012) 0.104*** (0.012) L3.Exports 0.104*** (0.010) 0.105*** (0.010) 0.100*** (0.011) 0.101*** (0.011) L4.Exports 0.123*** (0.009) 0.123*** (0.009) 0.117*** (0.010) 0.117*** (0.010) RER 0.120*** (0.029) 0.132*** (0.029) 0.155*** (0.032) 0.162*** (0.032) L.RER 0.022 (0.045) 0.015 (0.046) 0.009 (0.049) 0.009 (0.050) L2.RER -0.140*** (0.031) -0.088* (0.047) -0.151*** (0.034) -0.107** (0.052) L3.RER -0.036 (0.042) -0.051 (0.046) L4.RER -0.022 (0.027) 0.002 (0.030) REER 0.044*** (0.009) 0.041*** (0.009) 0.046*** (0.009) 0.044*** (0.010) L.REER -0.008 (0.010) -0.010 (0.010) -0.012 (0.010) -0.013 (0.010) L2.REER -0.013* (0.008) -0.016* (0.009) -0.012 (0.009) -0.013 (0.010) L3.REER -0.003 (0.009) -0.004 (0.010) L4.REER 0.009 (0.008) 0.007 (0.008) Demand 0.653*** (0.019) 0.648*** (0.020) 0.626*** (0.022) 0.625*** (0.022) L.Demand -0.148*** (0.025) -0.153*** (0.025) -0.103*** (0.028) -0.108*** (0.028) L2.Demand -0.113*** (0.024) -0.120*** (0.024) -0.109*** (0.027) -0.116*** (0.027) L3.Demand -0.098*** (0.023) -0.092*** (0.023) -0.101*** (0.026) -0.094*** (0.026) L4.Demand -0.098*** (0.018) -0.088*** (0.019) -0.093*** (0.020) -0.088*** (0.021) Uncertainty -0.004 (0.004) -0.004 (0.004) -0.009* (0.005) -0.009* (0.005) L.Uncertainty -0.013*** (0.004) -0.013*** (0.005) -0.013** (0.005) -0.013** (0.005) L2.Uncertainty -0.007 (0.005) -0.006 (0.005) -0.003 (0.005) -0.003 (0.005) L3.Uncertainty -0.007 (0.004) -0.004 (0.005) L4.Uncertainty -0.002 (0.004) -0.001 (0.005) Observations 73084 72850 59348 59114 R-squared 0.984 0.984 0.983 0.983 Note: Results of baseline estimations. The dependent variable in all columns is (log of) exports. All regressions include the exporter-year and exporter-importer �xed effects, coefficients not reported. Robust standard errors in parentheses. Signi�cance (p-value): *10%, **5%, ***1%. 29 Table 3: Baseline: Long run effects I. Baseline with 2 lags on shocks Variable Coefficient SD t-stat p-value Uncertainty -0.115 0.035 -3.30 0.001 Demand 0.948 0.033 28.69 0.000 RER 0.008 0.047 0.17 0.867 REER 0.107 0.026 4.09 0.000 II. Baseline with 4 lags on shocks Uncertainty -0.152 0.043 -3.49 0.000 Demand 0.944 0.033 28.30 0.000 RER 0.004 0.048 0.08 0.940 REER 0.107 0.027 3.97 0.000 III. Baseline with 2 lags on shocks pre 2007 sample Uncertainty -0.114 0.036 -3.12 0.002 Demand 0.971 0.039 24.82 0.000 RER 0.057 0.051 1.12 0.264 REER 0.095 0.027 3.49 0.000 IV. Baseline with 4 lags on shocks pre 2007 sample Uncertainty -0.134 0.045 -2.94 0.003 Demand 0.964 0.039 24.52 0.000 RER 0.063 0.052 1.23 0.220 REER 0.094 0.028 3.34 0.001 Note: Long run effects of all control variables on international trade, computed based on the results of baseline estimations in Table 2. Signi�cance (p-value): *10%, **5%, ***1%. 30 Table 4: MMR + Wealth (1) (2) (3) (4) (5) (6) Uncertainty -0.004 -0.005 -0.005 -0.012** -0.011** -0.013** (0.005) (0.004) (0.005) (0.005) (0.005) (0.005) L.Uncertainty -0.007 -0.011** -0.003 -0.005 -0.012** -0.001 (0.005) (0.004) (0.005) (0.005) (0.005) (0.005) L2.Uncertainty -0.008* -0.005 -0.006 -0.004 -0.002 -0.002 (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) MMR -0.000 0.000 -0.000 0.000 (0.000) (0.000) (0.000) (0.000) L.MMR -0.002*** -0.002*** -0.002*** -0.002*** (0.001) (0.001) (0.001) (0.001) L2.MMR 0.001*** 0.001*** 0.001*** 0.001*** (0.000) (0.000) (0.000) (0.000) Wealth 0.050*** 0.039** 0.057*** 0.045** (0.015) (0.016) (0.016) (0.018) L.Wealth 0.014 0.012 0.003 -0.004 (0.022) (0.025) (0.024) (0.027) L2.Wealth -0.059*** -0.048*** -0.054*** -0.039** (0.014) (0.016) (0.015) (0.018) Observations 71595 61432 57859 48130 R-squared 0.984 0.986 0.984 0.986 Note: Results of baseline regressions augmented by money market rate (MMR) and stock market adjustments (Wealth). The dependent variable in all columns is (log of) exports. The lagged dependent variable and a measure of demand are included up to lag four and exchange rates up to lag two, but coefficients are not reported. All regressions also include the exporter-year and exporter-importer �xed effects, coefficients not reported. Robust standard errors in parentheses. Signi�cance (p-value): *10%, **5%, ***1%. 31 Table 5: MMR + Wealth: Long run effects I. Baseline + MMR IV. Baseline + MMR pre 2007 Variable Coef SD t-stat p-val Variable Coeff SD t-stat p-val Uncertainty -0.076 0.034 -2.21 0.027 Uncertainty -0.077 0.037 -2.09 0.037 Demand 0.722 0.041 17.55 0.000 Demand 0.737 0.052 14.10 0.000 RER 0.106 0.055 1.93 0.054 RER 0.138 0.061 2.28 0.023 REER 0.111 0.030 3.66 0.000 REER 0.085 0.033 2.55 0.011 MMR -0.009 0.002 -4.79 0.000 MMR -0.008 0.002 -4.53 0.000 II. Baseline + wealth effect V. Baseline + wealth effect pre 2007 Uncertainty -0.105 0.034 -3.05 0.002 Uncertainty -0.104 0.036 -2.90 0.004 Demand 0.964 0.036 26.58 0.000 Demand 1.016 0.043 23.73 0.000 RER 0.038 0.047 0.80 0.423 RER 0.063 0.052 1.21 0.224 REER 0.100 0.027 3.73 0.000 REER 0.087 0.028 3.11 0.002 Wealth 0.024 0.013 1.83 0.067 Wealth 0.026 0.013 1.94 0.052 III. Baseline + MMR + wealth effect VI. Baseline + MMR + wealth effect pre 2007 Uncertainty -0.066 0.034 -1.93 0.054 Uncertainty -0.068 0.037 -1.85 0.064 Demand 0.704 0.043 16.44 0.000 Demand 0.722 0.053 13.54 0.000 RER 0.124 0.054 2.29 0.022 RER 0.144 0.060 2.40 0.016 REER 0.099 0.031 3.21 0.001 REER 0.070 0.034 2.06 0.039 MMR -0.005 0.001 -3.57 0.000 MMR -0.004 0.001 -3.27 0.001 Wealth 0.014 0.016 0.90 0.368 Wealth 0.011 0.015 0.71 0.477 Note: Long run effects of all control variables on international trade, computed based on the results of baseline estimations in Table 4. Signi�cance (p-value): *10%, **5%, ***1%. 32 Table 6: Non-linearity (1) (2) (3) (4) (5) (6) (7) (8) SD*0.5 SD*0.5 SD*1 SD*1 SD*1.5 SD*1.5 SD*2 SD*2 Uncertainty 0.000 -0.001 -0.001 -0.001 -0.006 -0.006 -0.011** -0.009* (0.003) (0.003) (0.004) (0.004) (0.004) (0.004) (0.005) (0.005) L.Uncertainty -0.008** -0.005 -0.009** -0.003 -0.009** -0.000 -0.018*** -0.007 (0.003) (0.003) (0.004) (0.004) (0.004) (0.004) (0.005) (0.005) L2.Uncertainty -0.000 0.002 -0.003 -0.001 -0.007 -0.006 -0.013*** -0.012** (0.003) (0.003) (0.004) (0.004) (0.004) (0.004) (0.005) (0.005) MMR 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) L.MMR -0.002*** -0.002*** -0.002*** -0.002*** (0.001) (0.001) (0.001) (0.001) L2.MMR 0.001*** 0.001*** 0.001*** 0.001*** (0.000) (0.000) (0.000) (0.000) Wealth 0.040** 0.040** 0.039** 0.036** (0.016) (0.016) (0.016) (0.016) L.Wealth 0.014 0.012 0.013 0.012 (0.025) (0.025) (0.025) (0.025) L2.Wealth -0.051*** -0.050*** -0.049*** -0.044*** (0.016) (0.016) (0.016) (0.016) Observations 71595 61432 71595 61432 71595 61432 71595 61432 R-squared 0.984 0.986 0.984 0.986 0.984 0.986 0.984 0.986 Note: Results based on different de�nitions of uncertainty thresholds. SD stands for standard deviation and the number in front refers to the threshold level that has been used to construct the Uncertainty measure. The dependent variable in all columns is (log of) exports. The lagged dependent variable and a measure of demand are included up to lag four and exchange rates up to lag two, but coefficients are not reported. All regressions also include the exporter-year and exporter-importer �xed effects, coefficients not reported. Robust standard errors in parentheses. Signi�cance (p-value): *10%, **5%, ***1%. 33 Table 7: Non-linearity: Long run effects I. Baseline, Shock = 0.5*SD V. Baseline, Shock = 1.5*SD Variable Coef SD t-stat p-val Variable Coeff SD t-stat p-val Uncertainty -0.040 0.023 -1.74 0.082 Uncertainty -0.104 0.032 -3.22 0.001 Demand 0.972 0.034 28.60 0.000 Demand 0.962 0.034 28.38 0.000 RER 0.017 0.047 0.35 0.724 RER 0.021 0.047 0.45 0.655 REER 0.108 0.026 4.12 0.000 REER 0.110 0.026 4.21 0.000 II. Baseline+MMR+Wealth, Shock = 0.5*SD VI. Baseline+MMR+Wealth, Shock = 1.5*SD Uncertainty -0.016 0.023 -0.68 0.493 Uncertainty -0.058 0.032 -1.83 0.068 Demand 0.710 0.043 16.60 0.000 Demand 0.704 0.043 16.41 0.000 RER 0.124 0.054 2.29 0.022 RER 0.125 0.054 2.32 0.020 REER 0.097 0.031 3.14 0.002 REER 0.099 0.031 3.21 0.001 MMR -0.005 0.001 -3.65 0.000 MMR -0.005 0.001 -3.60 0.000 Wealth 0.012 0.016 0.78 0.434 Wealth 0.014 0.016 0.87 0.385 III. Baseline, Shock = 1*SD VII. Baseline, Shock = 2*SD Uncertainty -0.065 0.028 -2.33 0.020 Uncertainty -0.205 0.040 -5.11 0.000 Demand 0.965 0.034 28.50 0.000 Demand 0.962 0.034 28.47 0.000 RER 0.018 0.047 0.39 0.697 RER 0.025 0.047 0.52 0.600 REER 0.109 0.026 4.17 0.000 REER 0.106 0.026 4.09 0.000 IV. Baseline+MMR+Wealth, Shock = 1*SD VIII. Baseline+MMR+Wealth, Shock = 2*SD Uncertainty -0.022 0.028 -0.77 0.444 Uncertainty -0.127 0.039 -3.24 0.001 Demand 0.708 0.043 16.54 0.000 Demand 0.708 0.043 16.56 0.000 RER 0.124 0.054 2.30 0.022 RER 0.122 0.054 2.26 0.024 REER 0.097 0.031 3.16 0.002 REER 0.099 0.031 3.21 0.001 MMR -0.005 0.001 -3.64 0.000 MMR -0.005 0.001 -3.41 0.001 Wealth 0.013 0.016 0.80 0.422 Wealth 0.016 0.016 0.99 0.321 Note: Long run effects of all control variables on international trade, computed based on the results of baseline estimations in Table 6. Signi�cance (p-value): *10%, **5%, ***1%. 34 Table 8: Durables (1) (2) (3) (4) Top quartile Third quartile Second quartile Bottom quartile Uncertainty -0.009 0.007 -0.015* -0.008 (0.009) (0.009) (0.009) (0.010) L.Uncertainty 0.005 -0.021** -0.015* -0.013 (0.010) (0.009) (0.009) (0.009) L2.Uncertainty 0.004 -0.020** -0.008 -0.001 (0.010) (0.009) (0.009) (0.010) Constant -1.213*** -1.215*** -0.853*** -0.654*** (0.175) (0.208) (0.148) (0.129) Observations 18988 19020 16470 18606 R-squared 0.982 0.985 0.988 0.987 Note: Impact of uncertainty at different levels of durability of a trading relationships. Trading relationships for each importers are split into quartiles by the share of durables traded with the top quartile referring to the highest share of durables. The dependent variable in all columns is (log of) exports. The lagged dependent variable and a measure of demand are included up to lag four and exchange rates up to lag two, but coefficients are not reported. All regressions also include the exporter-year and exporter-importer �xed effects, coefficients not reported. Robust standard errors in parentheses. Signi�cance (p-value): *10%, **5%, ***1%. 35 Table 9: Durables: Long run effects I. Durables: top quartile Variable Coefficient SD t-stat p-value Uncertainty 0.003 0.072 0.04 0.971 Demand 1.041 0.067 15.64 0.000 RER 0.084 0.095 0.89 0.375 REER 0.128 0.051 2.53 0.012 II. Durables: third quartile Uncertainty -0.142 0.060 -2.38 0.017 Demand 0.978 0.069 14.22 0.000 RER 0.032 0.080 0.40 0.690 REER 0.206 0.045 4.52 0.000 III. Durables: second quartile Uncertainty -0.205 0.076 -2.70 0.007 Demand 1.036 0.078 13.28 0.000 RER 0.183 0.111 1.65 0.099 REER 0.002 0.060 0.04 0.968 IV. Durables: bottom quartile Uncertainty -0.122 0.085 -1.43 0.153 Demand 0.860 0.066 13.10 0.000 RER -0.223 0.116 -1.93 0.053 REER 0.110 0.065 1.69 0.091 Note: Long run effects of all control variables on international trade, computed based on the results of baseline estimations in Table 8. Signi�cance (p-value): *10%, **5%, ***1%. 36 Table 10: Incidence of shocks: Long run effects (1) (2) (3) (4) (5) (6) Low Incidence High Incidence Experience Uncertainty 0.006 0.006 -0.003 -0.002 -0.004 -0.006 (0.007) (0.008) (0.006) (0.006) (0.004) (0.005) L.Uncertainty -0.012 -0.002 -0.016*** -0.007 -0.012*** -0.003 (0.007) (0.008) (0.006) (0.006) (0.005) (0.005) L2.Uncertainty -0.011 -0.011 -0.003 -0.002 -0.007 -0.005 (0.008) (0.009) (0.006) (0.006) (0.005) (0.005) MMR 0.001 -0.000 0.000 (0.001) (0.001) (0.000) L.MMR -0.005*** -0.002** -0.002*** (0.001) (0.001) (0.001) L2.MMR 0.003*** 0.001** 0.001*** (0.001) (0.001) (0.000) Wealth 0.058** 0.055** 0.038** (0.026) (0.021) (0.016) L.Wealth -0.040 0.001 0.012 (0.040) (0.031) (0.025) L2.Wealth -0.000 -0.055*** -0.047*** (0.026) (0.021) (0.016) Experience -0.001 -0.001 (0.001) (0.001) Observations 36098 30995 36986 30437 73084 61432 R-squared 0.985 0.987 0.983 0.986 0.984 0.986 Note: Impact of uncertainty on countries with low vs. high past incidence of shocks. Last two columns instead control for past experience with shocks explicitly. The dependent variable in all columns is (log of) exports. The lagged dependent variable and a measure of demand are included up to lag four and exchange rates up to lag two, but coefficients are not reported. All regressions also include the exporter-year and exporter-importer �xed effects, coefficients not reported. Robust standard errors in parentheses. Signi�cance (p-value): *10%, **5%, ***1%. 37 Table 11: Incidence of shocks: Long run effects I. Low incidence of shocks: baseline Variable Coefficient SD t-stat p-value Uncertainty -0.079 0.062 -1.28 0.199 Demand 1.018 0.044 23.19 0.000 RER -0.096 0.063 -1.52 0.129 REER 0.124 0.041 3.02 0.003 II. Low incidence of shocks: baseline + mmr + wealth Uncertainty -0.026 0.057 -0.45 0.656 Demand 0.603 0.065 9.31 0.000 RER 0.150 0.072 2.10 0.036 REER 0.128 0.040 3.19 0.001 MMR -0.007 0.002 -2.92 0.004 Wealth 0.077 0.031 2.46 0.014 III. High incidence of shocks: baseline Uncertainty -0.108 0.045 -2.40 0.017 Demand 0.816 0.050 16.18 0.000 RER 0.175 0.076 2.29 0.022 REER 0.127 0.036 3.53 0.000 IV. High incidence of shocks: baseline + mmr + wealth Uncertainty -0.052 0.046 -1.14 0.256 Demand 0.734 0.065 11.26 0.000 RER 0.052 0.091 0.58 0.564 REER 0.081 0.049 1.63 0.103 MMR -0.005 0.002 -2.86 0.004 Wealth 0.000 0.019 0.02 0.987 V. Controlling for experience with shocks: baseline Uncertainty -0.114 0.035 -3.26 0.001 Demand 0.946 0.033 28.44 0.000 RER 0.011 0.047 0.23 0.816 REER 0.107 0.026 4.10 0.000 VI. Controlling for experience with shocks: baseline + mmr + wealth Uncertainty -0.064 0.034 -1.87 0.061 Demand 0.695 0.044 15.95 0.000 RER 0.130 0.054 2.39 0.017 REER 0.098 0.031 3.17 0.002 MMR -0.005 0.001 -3.65 0.000 Wealth 0.015 0.016 0.98 0.325 Note: Long run effects of all control variables on international trade, computed based on the results of baseline estimations in Table 10. Signi�cance (p-value): *10%, **5%, ***1%. 38