2.1. Data

Our main source of data is S&P Capital IQ (CIQ), which provides detailed firm balance sheet and income statement information. CIQ has two key advantages compared with other leading corporate data providers such as Orbis or Worldscope. First, the data are available at the quarterly frequency, which is more suited to identify the firm-level responses to high frequency shocks—such as monetary policy shocks. Second, CIQ contains information on foreign-currency liabilities which allows us to quantify the importance of the financial channel of the exchange rate. The main limitation of the data is that the coverage of non-listed firms is insufficient in many countries. To maximize the consistency and reliability of the data, we therefore restrict our sample to listed firms —both active and inactive. In addition, we do not observe which financial institutions firms borrow from and are therefore not able to control for credit constraints arising from the supply side.

Our dataset covers a long time span and a broad set of countries—20 years of data, from 1996Q3 to 2016Q3, for 63 countries (29 AEs and 34 EMDEs). Firms in our dataset belong to a wide range of industries—20 CIQ-defined industries in total, after filtering out firms in the financial, insurance and utilities sectors. Details on the distribution of firms across countries and sectors are shown in Figures A2.1 and A2.2. Data are collected by CIQ on a consolidated basis only. To avoid double counting, we keep only companies that are ultimate corporate parents. We also exclude state-owned firms, by dropping them from the sample if the type of corporate parent is identified as a government body. To clean the data, we mainly follow Kim (2019) and Kim et al. (2020). Firms with negative total assets or total debt in a given year are entirely dropped from the sample, while firm-observations with unexpected signs for capital expenditure, for net property plant and equipment (NPPE), and for revenue are also excluded. In addition, an observation is filtered out if the difference between assets and liabilities is greater than USD 10,000, or if the amount of cash & cash equivalents and that of tangible assets are greater than total assets. All variables are winsorized at 5 percent to exclude outliers. After filtering, the sample consists of 23,482 firms.⁴ The panel, however, is highly unbalanced. The size of the economy and different filing requirements across countries also introduces large disparities in firm coverage across countries.

Our main variable of interest, the investment rate, is defined as the ratio of capital expenditures to NPPE. To cross-validate our firm-level investment data, we compare it with investment data at the macro level. We do this for the log-level of investment as well as its growth rate. In Table A2.3, we report the estimation results from regressing the log-investment calculated by aggregating our firm-level data against the log-investment from macro data. The results show that the level of investment is highly correlated (R-squared above 0.7) across the two sources. In addition, we find that the relation between these variables remains strong and highly statistically significant when we control for country fixed effects, and when considering growth rates.

To identify the role of the financial channel of the exchange rate, we define the foreign-currency liability ratio as foreign-currency liabilities to total liabilities. Foreign-currency liabilities are computed using data from CIQ’s Capital Structure module and, unlike the vast majority of studies, cover both bank debt and bond issuance.⁵ It is worth noting that we do not observe the extent to which firms hedge their foreign-currency exposures. However, a few studies suggest that foreign exchange hedging practices are not widespread in EMDEs (for example, Chui et al. 2014; Chuaprapaisilp et al. 2018) and, if undertaken, their effect would in fact work against finding a significant effect.

Data limitations prevent us from exploiting firm-level heterogeneity when it comes to the trade channel. Although CIQ provides export revenue data for firms, the coverage is very limited. To obviate this data limitation, in the baseline we use the World Input Output Database (WIOD), which provides cross-country trade data for 56 sectors in 43 countries between 2000 and 2014.⁶ We calculate the average export dependence for each sector-country pair during this period such that:

where c and s denote country and sector, respectively. ExportsDependence_cs denotes the average share of exports (Exports_cs) in sector s and country c in the value of total output (Output_cs).⁷ In the robustness check section, we repeat the analysis using a firm-level measure of export dependence and find that the results are qualitatively the same.

The regressions also include several additional firm-level characteristics: leverage, liquidity, size, and collateral. The definitions of all variables and their summary statistics are provided in Table 1.

2.2. U.S. Monetary Policy Shocks

We follow Duval et al. (2021) and Albrizio et al. (2021) in order to identify exogenous U.S. monetary policy shocks. In particular, we rely on high-frequency monetary policy surprises—that is, changes in Federal fund futures around FOMC announcements within a tight window of 30 minutes following Gürkaynak et al. (2005)—as a measure for unexpected monetary policy shocks. Here, the identification assumption is that the response of agents in financial markets reflects exclusively monetary policy news during this time interval. Following Gertler and Karadi (2015), we use these surprises in a proxy-SVAR framework to instrument one-year government bond yield together with industrial production, consumer price index and a measure of the excess bond premium from Gilchrist and Zakrajsek (2012).

We estimate this 4-variable proxy-SVAR over the period 1973M1–2016M8 at a monthly frequency to retrieve the structural monetary policy shocks (Ramey 2016).⁸ By using the one-year government bond yield, we are able to capture the impact of forward guidance to a larger extent than using simply the Fed funds rate, which had remained constant and close to zero during the post-global financial crisis period. In addition, we allow for a structural break in the VAR coefficients to account for the post zero-lower bound period and the beginning of quantitative easing in November 2008. Since the firm-level dataset is at a quarterly frequency, we aggregate the series of structural shocks within each quarter (see Figure A2.3 for the evolution of the estimated monetary policy shocks).

2.3. Empirical Strategy

Our empirical approach to quantifying U.S. monetary spillovers consists of three steps. In the first step, we estimate the average (unconditional) effect of U.S. monetary policy shocks on firm investment using Jorda’s (2005) local projections. Specifically, we estimate the following specification:

where the dependent variable, Y_ft, is the investment ratio or revenue of firm f at time (quarter) t; $i_t^{US}$ denotes the exogenous U.S. monetary policy shock at time t; γ_f indicates firm fixed effects to control for unobservable time-invariant firm characteristics; and γ_csq are country-sector-quarter dummies to account for cross-sector variations across countries as well as seasonality in the data.

We then expand Equation (1) to estimate how the effect of U.S. monetary policy shocks varies across firms according to the transmission channels discussed earlier. We follow the semi-parametric approach of Cloyne et al. (2019) and Duval et al. (2021), which allows us to estimate the impact of U.S. monetary policy shocks in a more flexible way, without making assumptions about the functional form of the relationship between firms’ characteristics and spillovers. In particular, we estimate the following specification:

where I is an indicator function which equals one if the observation associated with the firm characteristic X_f falls within a specific group g. For example, if the average leverage of a firm is above (below) the 75^th (25^th) percentile of average leverage across all firms, then the firm is classified in the “high leverage” (“low leverage”) group. We use the average over time of the firm characteristics to reduce endogeneity due to the potential time-varying response of firm characteristics to monetary policy shocks. γ_cst+h are country-sector-time fixed effects to account for macro-economic shocks and their differential effects across sectors (such as, for example, the sectoral differential effect of monetary policy shocks, both domestic and originating from the U.S.) as well as sector-specific shocks at the country level (such as, for example, changes in country regulations affecting a given sector). Z_ft-1 are firm-specific characteristics (leverage, liquidity, size and collateral) potentially affecting firm investment or revenues, lagged by one period to reduce reverse causality concerns; and the other terms are as described in Equation (1). We estimate Equation (2), first, separately for each transmission channel, and then considering all three channels together.⁹

Finally, we are also interested in examining whether the differential response related to a given transmission channel (for example, the balance sheet channel) varies with other firm characteristics, including those that relate to another transmission channel (for example, foreign-currency liabilities). To this end, we estimate the following equation:

where the indicator function, I, uses two criteria that are interacted with each other, for example, low/high leverage and small/medium/large firms. Note that this approach also allows us to further validate the importance of a given channel. For example, if we observe a statistically significant differential response in investment between low and high leverage firms for different levels of foreign-currency liabilities, we can conclude that leverage is a spillover transmission mechanism independent of foreign-currency liabilities.

Equations (1)-(3) are estimated using OLS and standard errors are two-way clustered on firm and country-time.

3.1. Average Effects

Figure 1 shows the average investment response to a 25-basis points U.S. monetary policy tightening. As expected, a U.S. monetary policy tightening is followed by an economically and statistically significant decline in foreign firms’ investment, with a peak impact of -1.3 percent after two quarters. The estimated investment response of foreign firms is comparable to, albeit slightly smaller than, the response of the U.S. firms in our sample (Figure A3.1). The effect is also similar to estimates in other studies using similar data, for example, Jeenas (2019) and Cloyne et al. (2019).

We conduct a range of robustness checks to confirm that our estimates are not driven by specific countries and are robust to alternative specifications. In particular, the results are robust to removing countries with the highest number of firms among AEs and EMDEs, one at a time (Figure A3.2).¹⁰ The results are also robust to alternative specifications, including: alternative lags of the dependent variable (Figure A3.4), alternative measures of monetary policy shocks (Figure A3.5), the addition of time-varying firm characteristics—such as leverage, collateral, liquidity and size—as controls (Figure A3.6), alternative measures of investment—such as the change in capital expenditures (Figure A3.7) or the ratio of capital expenditures to total assets (Figure A3.8) . Finally, we also find that U.S. monetary policy shocks have sizeable spillovers on firm revenue— a 25-basis points U.S. monetary policy tightening leads to a peak decline in revenue of about 0.6 percent after two quarters (Figure A3.9).

3.2. Spillover Channels and Firm Characteristics

In this subsection, we explore the role of the three key transmission channels—the balance sheet channel, the financial channel of the exchange rate, and the trade channel—by considering the differential responses of firms having different degrees of exposure to these channels. We assign firms into different exposure groups (low/medium/high) by using the cross-country distribution of firms with respect to different proxy variables. As discussed earlier, all firms with an average exposure variable below (above) the 25^th (75^th) percentile are classified into the low (high) group.¹¹ We begin our analysis by focusing on the role of the firm balance sheet channel, using firm leverage as the main exposure variable.¹² Figure 2 displays the differential investment response between firms in the high leverage group and the low leverage group (β_{HighLeverage,h} – β_{LowLeverage,h}) following a 25-basis points U.S. monetary policy shock, estimated using Equation (2). Consistent with the presence of a significant balance sheet channel, the results suggest that highly leveraged firms are more responsive to U.S. monetary policy shocks than firms with low leverage. The differential impulse responses are statistically significant over most of the estimation horizons, suggesting a persistent differential effect. The effects are quantitatively large: the difference in the one-year-ahead response of investment to a 25-basis points monetary policy shock between the top and bottom quartiles of the firm leverage distribution is about 0.5 percentage point. Next, we look at the role of the financial channel of the exchange rate by focusing on firms’ foreign-currency debt to total debt ratio (foreign-currency liability ratio). Figure 3 presents the differential impulse responses between firms with high and low foreign-currency liability ratios (β_HighFX,h – β_LowFX,h). The differential response is statistically significant, suggesting a larger decline in investment for firms that have a higher foreign-currency liability ratio. The magnitude of the differential response between the top and bottom quartiles is about 0.5 percentage point and is of similar size to that obtained with leverage.

Finally, we examine the role of the trade channel, using the export dependence of a sector as a proxy. Unlike the balance sheet channel and the financial channel of the exchange rate, the trade channel has an ambiguous spillover effect on foreign firms. On the one hand, a U.S. monetary policy tightening decreases the demand for domestic and foreign goods, negatively impacting firms operating in sectors with higher export dependence (expenditure reducing channel). On the other hand, an increase in the U.S. interest rate depreciates local currencies against the U.S. dollar, resulting in an increase in competitiveness (expenditure switching channel). Figure 4 shows the differential impulse response to a U.S. monetary policy shock for the high and low export dependence groups (β_{HighDependence,h} – β_{LowDependence,h}). Our results imply that both channels are at play. While in the short term, firms operating in sectors with higher export dependence are more negatively affected than firms operating in sectors with lower export dependence, the opposite holds about two years after the U.S. monetary policy shock. In other words, while the expenditure reducing channel seems to dominate in the short term, the expenditure switching channel dominates in the medium term.¹³ This seems consistent with the lagged effect of exchange rate fluctuations on exports, as found in the literature (for example, Leigh et al. 2017).

3.3. Robustness Checks

To check the robustness of our results, we perform a number of sensitivity tests across alternative samples and specifications. First, to check whether each channel plays a distinct and independent role in the transmission of U.S. monetary policy shocks, we extend Equation (2) by including all three channels together in the same specification:

where Lev_f, FX_f, and Trade_f denote the average leverage ratio, foreign-currency liability ratio, and export dependence ratio of the firm.¹⁴

Figure 5 displays the differential impulse responses for each channel from estimating Equation (4). The results are qualitatively similar to, and not statistically different from, those obtained from estimating each channel separately, suggesting that each channel has an independent and economically and statistically significant role in shaping spillovers from U.S. monetary policy.

Second, we expand the set of control variables to include the interactions of monetary policy shocks with other firm characteristics capturing the firm’s financial strength—such as firm size, bank debt to total debt ratio, dividend payment status and firm age. While we do find that these factors contribute to amplify the spillover effects of U.S. monetary policy shocks (Figure A4.1)—with smaller, younger firms paying lower dividends experiencing a larger decline in investment—their inclusion as control variables in our specification does not significantly affect the results for the differential impact of leverage, foreign-currency liability and trade exposure (Figure A4.2).

Third, we show that our results are robust to alternative methods for classifying firms into different groups. We find similar results if we use the country-specific distribution instead of the full cross-country distribution to determine the thresholds for classifying firms into low/medium/high groups (Figure A4.3) or if we only consider a low/high group based on the median of the distribution (Figure A4.4).

A concern regarding our firm classification approach is the fact that our exposure variables may be endogenous to monetary policy shocks. For example, firms might become more leveraged when interest rates decrease in the United States or foreign firms might find it cheaper to borrow in foreign currency when U.S. monetary policy is more accommodative. To address this concern, we classify firms using their characteristics from the first year they report data, as opposed to the average over the full sample period. Figure A4.5 shows that the significance of each channel is robust to this strategy.

Next, to show that the significance of the spillover channels discussed in the text is not driven by any particular country, we drop from our sample, one at a time, the six countries with the highest number of firms (Canada, China, Hong Kong, India, Japan, Taiwan) while estimating Equation (2). The results are similar to, and not statistically different from, those obtained in the baseline (Figure A4.6).

Finally, we examine the response of revenue through these channels. The results suggest that, as expected, the trade channel plays a larger role for revenue than for investment, while the opposite is true for the two other channels (Figure A4.7).

3.4. Back-of-the-Envelope Calculations

Our results suggest a statistically significant role for all three channels, but how important are these channels in terms of explaining the aggregate investment response to U.S. monetary policy shocks? To answer this question, we follow Ciminelli et al. (2020); we quantify the contribution of each channel to the average investment response by using our estimates of Equation (4) and calculate the contribution of each channel.

For any given channel, the average response of investment can be written as the weighted average of investment for the three groups (for example, high/medium/low leverage):

where K denotes investment, ω_i and $K_t^i$ are the capital expenditure share and investment of firms in group i, respectively. Taking the partial derivative of the equation above and using the fact that ω₃ = 1 – ω₁-ω₂ yields the following expression:

From Equation (4) we have estimates of $(\frac{\partial K_t^1}{\partial i_t^{US}}-\frac{\partial K_t^3}{\partial i_t^{US}})$ for each channel. Assuming that $\frac{\partial K_t^3}{\partial i_t^{US}}=0$ and $(\frac{\partial K_t^2}{\partial i_t^{US}}-\frac{\partial K_t^3}{\partial i_t^{US}})$ yields the following expression for the contribution of a given channel to the overall investment response: $\left({\omega }_1\frac{\frac{\partial K_t^1}{\partial i_t^{US}}}{\frac{\partial K_t}{\partial i_t^{US}}}\right)$. For $\frac{\partial K_t}{\partial i_t^{US}}$, we rely on our estimates of Equation (1) and calculate the contribution of each channel accordingly. Note that by assuming that $\frac{\partial K_t^2}{\partial i_t^{US}}=0$ and $\frac{\partial K_t^3}{\partial i_t^{US}}=0$, we significantly under-estimate the contribution of each channel to the average investment response. Therefore, the results of this exercise, while useful to examine the relative strength of each channel, should be considered as a lower bound of the true general equilibrium effect of each channel.

Table 2 reports the cumulative (unconditional and conditional) investment response to U.S. monetary policy shocks and the contribution of each channel. The results suggest that the balance sheet channel accounts for the largest share of the total investment response, followed by the exchange rate and trade channels. In contrast, consistent with the estimation results for revenue, back-of-the-envelope calculations suggest that the trade channel accounts for the largest share of the total revenue response (Table A4.1).

3.5. Interactions Between Channels

In this subsection, we explore potential interactions between channels using the cross-classifications described in Equation (3), where firms are partitioned along multiple firm characteristics capturing different channels.

We first consider the interaction between the balance sheet channel and the financial channel of the exchange rate. To do so, we first partition the sample conditional on the firms’ foreign-currency liability ratio and consider the differential response of firms with low versus high leverage (Figure 6 top panel); and then partition the sample conditional on firms’ leverage and consider the differential response between firms with a high and low foreign-currency liability ratio (Figure 6 bottom panel). The results in Figure 6 suggest that both the leverage and foreign-currency liability ratio have a significant differential impact on firm investment, irrespective of each other. As expected, we also find that the transmission channel through higher leverage is amplified by a higher foreign-currency exposure.¹⁵

Next, we explore the interaction between the balance sheet and the trade channels. In the upper panels of Figure 7, we present the differential impulse responses between high and low leverage firms after splitting the sample into groups: firms with low (left panel) and high (right panel) export dependence ratios. The results show that leverage has an important role, independent of the firms’ export dependence ratio. The bottom panels in Figure 7 show the role of the export dependence ratio among firms with low and high leverage. This exercise shows that the role of the export dependence ratio is not statistically significant after firms are grouped based on their leverage—suggesting that trade exposure is a less robust channel of monetary spillovers on firm investment. On the other hand, Figure A4.9 suggests that the role of trade is important and statistically significant on firm revenue irrespective of the leverage group of the firm.

The next exercise examines the interaction between the financial channel of the exchange rate and the trade channel. Figure 8 shows the difference in responses between firms with a high versus low foreign-currency liability ratio, among low versus high export dependence ratio groups. The results suggest a significant role for the financial channel of the exchange rate for firms operating in sectors with high export dependence, but the foreign-currency debt share has no statistically significant role among firms in the low export dependence group. Figure 8 bottom panel displays the role of the trade channel among low and high foreign-currency liability ratio firms. As for leverage, the results from this exercise suggest that the trade channel does not have a significant differential impact on investment after firms are grouped based on their foreign-currency liability ratios. At the same time, for firm revenue we find a significant role for trade exposure independent of the firms’ foreign-currency liability ratio group (Figure A4.10).

3.6. Extensions

In this section, we consider additional potential interactions. First, we consider the role of other firm characteristics, for example, firm size and liquidity, which may amplify the role of leverage in the transmission of U.S. monetary policy spillovers. Using cross-classifications described in Equation (3), we group firms along two dimensions, first by leverage and size, and then by leverage and firm liquidity. Figure A4.11 top and bottom panels present the differential impulse responses for the leverage ratio, conditional on size and liquidity position (left versus right chart), respectively. Our results suggest that the impact of leverage is more pronounced for smaller and less liquid firms, consistent with financial frictions being more binding for smaller and less liquid firms.

To further investigate the role of the financial channel of the exchange rate, we look at how our estimates depend on a country’s exchange rate regime. In other words, we partition firms based on their foreign-currency liability ratio and the exchange rate regime (flexible vs. fixed) of the country where they are located. As expected, a firm’s foreign-currency borrowing has a more pronounced impact on its balance sheet and investment in response to U.S. monetary policy shocks in countries with more flexible exchange rate regimes (Figure A4.12).