For identification of the US interest rate shocks, we use the monthly dataset from 2007 to 2020 on the Industrial Production Index (IPI), Consumer Price Index (CPI), securities held outright by the Federal Reserve, and shadow interest rates of the United States. As mentioned before, we take the shadow rate as an alternative to the effective FFR to incorporate the effects of QE. To account for the zero lower bound and the stimulus provided by the unconventional monetary policy measures that followed the GFC, we employ Wu-Xia (2016) shadow rate for the US interest rate. Their shadow rate estimates are substantially associated with the QE-related asset holdings by the Federal Reserve, as shown by Wu and Zhang (2019), who define the shadow rate as the goal that the Federal Reserve hopes to achieve through bond purchases and sales.

We use a monthly dataset from 2007 to 2020 for a total of 29 emerging economies to examine the spillover effects of US interest rate shocks. The Appendix section contains the list of the nations that were considered for the analysis. The dataset consists of each EME’s IPI, CPI, exchange rate, money supply (broad money), and foreign exchange reserves. We use the global database platform CEIC for collecting the data on all the variables. We take the natural log of the variables in the study. The Appendices A–D provides a comprehensive explanation of the creation of the dataset.

To get a more accurate picture of the economic conditions of the EMEs, we incorporate the IPI and CPI in our model. Our focus, however, is on the response of broad money, exchange rates, and total reserves in the emerging markets. The inclusion of emerging market countries in the sample is driven by the availability of comprehensive monthly data. We try to incorporate a wide range of countries from various regions of the world.

Fig. 1 depicts the effective FFR, and the shadow rate provided by Wu-Xia. Following the GFC in 2007, the Federal Reserve in the United States quickly reduced interest rates to just above zero percent. Therefore, for further stimulation of the economy, it resorted to unconventional measures such as large-scale asset purchases, widely known as quantitative easing. The shadow rate captures these effects and can be below the zero lower bound as seen in Fig. 1.

Fig. 1.

The monthly effective federal fund rate and Wu-Xia shadow rates, 2007–2020 (%).

Note: In the periods from 2009 to 2016, the shadow rate fell beyond the zero lower bound to capture the stimulus from unconventional monetary policy measures in the US. Source: Compiled by the author.

2.2.1. Identification of US interest rate shocks

In order to study the effects of US interest rates, we need to examine the reasons behind the changes in interest rates since these causes may have a varied effect on EMEs. The causes can be investigated with the use of the Taylor rule. According to it, the interest rate is set at r = f (z) + u, where r is the US interest rate or shadow rate, z is the state of US economy, and u is the monetary policy shock. US interest rates may rise because of z (a strong US economy) or u (monetary policy shocks). Hence, if, for instance, the US interest rate rises due to a monetary policy shock, these unanticipated changes in the US interest rate might result in an outbound capital flow from the EMEs. A different response, however, might occur if the US interest rate increases because the US economy is stable. Basically, a robust US economy may encourage investors to be less risk averse, which would increase capital flows to EMEs (Rey, 2015). The focus of this study is to look at how these unanticipated changes brought on by monetary policy shocks affect the EMEs. To comprehend these effects, we use a US structural VAR (SVAR) model, in which the z variable is a collection of factors representing the stability of the US economy and the residual term, i.e., u, represents the interest rate shocks, as follows:

r_t = α ₀ + α₁ z_t + u_t, (1)

where r_t is the US interest rate, z_t includes contemporaneous and lagged values­ of log IPI, log CPI, log of securities held outright on the balance sheet of the Federal Reserve and lagged values of interest rates¹. We estimate the shocks as the unexpected changes in interest rates that are not the result of the stability of the US economy. Our structural identification strategy imposes a Cholesky decomposition of the covariance matrix. In the Cholesky identification criteria, the first factor does not respond to any other variable contemporaneously, the second factor only responds to contemporaneous changes in the first factor, and so on. Although all variables respond to lagged changes in each other. In our SVAR model, we order the shadow rates at the last. The SVAR model includes 6 lags of endogenous variables.

To examine the impact of US interest rates on the emerging markets, we use a panel SVAR model. The PSVAR model in our analysis differs from the general specifications of Pedroni in that we first do not demean the endogenous variables in levels and then take their first difference. Instead, we just take the first difference without centralizing the variables. The rationale behind this is that demeaning the endogenous variables does not have any impact on the first differences anyway. So basically, it does not make sense to first demean the endogenous variables in level and then take the differences. Secondly, we consider a mix of panel data sets and pure time series data. The US interest rate shock in our dataset is not a panel variable but a time series one since it does not change between the cross-section units of our analysis, i.e., countries.

Now, consider a panel composed of, i = 1, ..., N countries, each of which consists of M × 1 vector of observed endogenous variables, y_it, for y_m,it, m = 1, …, M. We will now take the stationary form to be in terms of the differences of the variables, namely y_i,t for facilitating the short-run restrictions on the dynamics. So, we use the following estimation model for our panel data set:

B_i y_i,t = A_i (L) y_i,t–1 + ε_i,t, (2)

where $A_i\left(L\right)=\sum _{s=0}^{s_i}{A}_{i,s}{L}^s$ is a lag polynomial allowing for country-specific lag lengths according to the usual information criteria, i = 1, …, N_t and t = 1, …, T_i; the i and t subscripts on the time and cross-section dimensions take into account that the panel may be unbalanced; B_i is the coefficient matrix; ε_i,t is the composite white noise shocks for M × 1 vector of endogenous variables, ε_m,it, m = 1, …, M.

The PSVAR by Pedroni (2013) distinguish the composite shocks into common shocks and idiosyncratic shocks. Hence, the common shocks in our model will be: ${\epsilon }_{i,t}={({\mathrm{\epsilon ̄}}_{it}^{\text{'}},{\mathrm{\epsilon ̃}}_{it}^{\text{'}})}^{\text{'}}, where {\mathrm{\epsilon ̄}}_{it}^{\text{'}} and {\mathrm{\epsilon ̃}}_{it}^{\text{'}}$ are M × 1 vectors of common and idiosyncratic white noise shocks respectively. Let Λ_i be an M × M diagonal matrix with loading coefficients Λ_i, m = 1, …, M. Then,

ε_i,t = Λ_i ε̄_it + ε̃̃_it, (3)

As mentioned earlier, our model differs from the general PSVAR model since we have US interest rates as pure time series data. Hence, the vector of endogenous variables i.e., y_it = (y_1,it, y_2,it)' is a mix of panel data and pure time series data. This means that for y_2,it, the cross-sectional average is trivially equal to itself. This makes it logically impossible to use it to identify common and idiosyncratic shocks.

The y_1,it panel contains the log IPI, log CPI, log exchange rate, log broad money, and log foreign exchange reserves. We follow Cholesky identification criteria by ordering the shadow rate shocks first so that they do not respond to the contemporaneous changes in the factors of the EMEs. The y_2,it, representing the US interest rate shocks, are the residuals that are estimated from the US SVAR model.

To obtain the structural residuals and responses, we estimate a set of N reduced-form VARs, one for each country i:

$∆y_{1,t}=B_1^{-1}{A}_1\left(L\right)y_{1,t-1}+{\epsilon }_{i,t`}$

$⋮$

$∆y_{N,t}=B_N^{-1}{A}_N\left(L\right)∆y_{N,t-1}+{\epsilon }_{N,t^{.}}$ (4)

Fig. 2 depicts the US interest rate shocks identified from the US SVAR model. In the period during 2008, we see the largest contractionary shock which is followed by another shock in 2009. We can also see the shock during the taper tantrum episode in the year 2013. Another episode of a contractionary shock was during the COVID 2019 pandemic.

Fig. 2.

Plots the unanticipated shocks in the US shadow interest rate, 2007–2020 (%).

Note: The unanticipated shocks are identified from the US SVAR model. The residuals from this model are considered as unanticipated shocks in the shadow rates. The largest contractionary shocks can be seen in 2008, depicting the quantitative easing measures. Source: Compiled by the author.

We are interested in the implications of US interest rates on the economic conditions of the emerging markets. The results from the PSVAR model are reported as impulse response functions (IRFs) in the following panels. The IRFs show how a one-standard deviation shock in one factor affects the other factor and how the effect dissipates over a course of time.

As mentioned earlier, since the US interest rate shock is a pure time-series variable, the idiosyncratic shocks are not relevant as the average effect will be somewhat the same. So, we only pay attention to the common shocks. The following Impulse Response Functions (IRF) show the point-wise median as well as the 25^th quantile (as Q1) and 75^th quantile (as Q2) confidence intervals (inter-quantile range) for the response functions of IPI, CPI, exchange rates, broad money, and foreign exchange reserves. It displays the responses of these factors to one-standard deviation shocks.

Fig. 3 shows the impulse responses of the macroeconomic factors of EMEs to a positive shock in the shadow rate. We estimate an immediate negative effect on IPI and an immediate positive effect on CPI. The exchange rates of the EMEs depreciate against the US dollar, indicating capital flow outbound from them. Moreover, the broad money supply and the foreign exchange reserves both decline immediately as a response to a positive shock in the shadow rate. This points to a domestic capital outflow from emerging markets. According to our findings, a positive shock in US interest rates has a contractionary effect on the EMEs. The estimates of responses to a negative change will be symmetrical to those of a positive shock.

Fig. 3.

Impulse response functions to shock in US shadow rates.

Note: The figure depicts the impulse response functions of EMEs to a one-standard deviation shock (positive shock) in the US interest shadow rates. The horizontal axis represents the time period or horizon, while the vertical axis represents the percent of responses. The range between Q3 and Q1 is the interquartile range. The point-wise median impulse responses represent the impact of a positive shock on the IPI, CPI, exchange rate, broad money (M3), and foreign exchange reserves, respectively. The IRFs show how these factors behave over a period of time in accordance with the shock in the US shadow rate. Source: Compiled by the author.

We now move to focus on the extent of these responses. The exchange rate depreciates by approximately 0.3% in period 3, with one standard deviation positive shock in the shadow rate. Broad money amounts to 0.41% in period 7. The effect on foreign exchange reserves declines to about 0.2% on the impact and finally it hits the negative peak at 0.4% in period 6. These responses reflect capital flight from emerging markets and demonstrate that interest rate shocks in the US create a safe haven situation for investors, causing them to shift to investing in a strong global currency, resulting in the depreciation of EME domestic currencies against the US dollar, a low monetary aggregate, and lower foreign exchange reserves.

In this section, we extend our analysis by addressing the bias in our dataset. The dataset is heavily skewed towards European countries, so the results can be biased in showing the full picture for non-European countries. Hence, we segregate the economies based on their regions, namely, Asia, Europe, and Latin America. The selection of regions is purely based on the number of countries in that region for which the data is available. The rationale behind this is to check if the results deviate across regions and from our earlier analysis as well. Due to the insufficient country count for the Middle Eastern and Sub-Saharan African regions, we do not divide them into sub-parts.

The results for the Asian countries are presented first. Fig. 4 shows the responses to a positive shock in the US shadow rate. Overall, the effects are contractionary in the Asian economies. However, the extent of these responses seems to be larger for foreign exchange reserves and, more importantly, they are sustained for a longer time period. The exchange rate depreciates to 0.3% in period 8, broad money dips to 0.3%, and foreign exchange reserves hit the lowest at –1%. From the IRFs for European economies, the depreciation in exchange rates of domestic currencies against the US dollar is about 0.5% (as seen in Fig. 5). Broad money dips to around 0.45% in period 7. Finally, the foreign exchange reserves fell by 0.55% in period 7. However, the response results from the Latin American economies differ from the above discussed responses. The estimate of the exchange rate shows that it appreciates as an immediate response to a positive shock. It suggests that the responses are rather delayed as the exchange rate depreciates in the later periods (as seen in Fig. 6). The reason for such a deviation might be due to the low number of countries in the region, which prevents us from seeing a rather full picture. Moreover, the responses of broad money and foreign exchange reserves are contractionary, consistent with the former results.

Fig. 4.

Impulse response functions to shock in US shadow rate: Asia.

Note: The figure depicts the IRFs to a one-standard deviation shock in US shadow rates for the monthly dataset on EMEs from Asia. The median impulse responses represent the percentage response of an unanticipated positive shock in US shadow rate on the Asian emerging markets over time. Source: Compiled by the author.

Fig. 5.

Impulse response functions to shock in US shadow rates: Europe.

Note: The figure depicts the IRFs to a one-standard deviation positive shock in the US shadow rates for a monthly dataset on EMEs from Europe. The point-wise median responses show the percentage response of macroeconomic factors of European emerging markets due to a positive shock in the US shadow rates over time. Source: Compiled by the author.

Fig. 6.

Impulse response functions to shock in US shadow rates: Latin America.

Note: The figure depicts the IRFs to a one-standard deviation positive shock in US shadow rates for a monthly dataset on EMEs from Latin America. The point-wise median responses are the percentage response of macroeconomic factors due to an unanticipated positive shock in the US shadow rate over time. Source: Compiled by the author.

The analysis by dividing the countries in accordance with their regions helps us to see the heterogenous effects on a particular region. The results make it absolutely clear that the interest rates in the US affect the money supply and foreign exchange reserves to decline. It suggests a capital outflow from the emerging markets due to such shocks.

Is using the shadow rate for identifying US interest rate shocks really effective? What difference does it make if we use effective FFR to identify US monetary shocks? We assess these by incorporating effective FFR instead of shadow rate in this segment in the US SVAR model. We then consider the residuals as the unanticipated shocks in the US interest rates and use them in the Panel SVAR by following the same procedure. The baseline specifications for both the US SVAR and Panel SVAR models are exactly the same as in the former model. The estimates from this model are presented in the Appendix section. The responses from this approach substantially differ from our earlier results. According to the IRFs, the exchange rate appreciates on the impact, and broad money as well as foreign exchange reserves rise immediately. However, the exchange rate starts depreciating from period 2. Similarly, broad money estimate also shifts direction and declines from period 2 onwards. Finally, the foreign exchange reserves decline from the impact value, although they do not decline below the zero percent point. These results convey the shock responses to be lagged, that is, we see the effects of unanticipated shocks after period 2 and onwards. This comparison highlights the importance of using the shadow rate to identify shocks because the effective FFR cannot fall below zero, so the effects of monetary stimulus would not be reflected in the FFR due to the liquidity trap. Hence, we believe it might not capture the full picture.