1.3 Data and empirical methodology
1.4.3 Sensitivity to alternative specifications
Next, we evaluate the sensitivity of the results from the panel dimension to changes in the specification of the model.21 We start by assessing the sensitivity of the results
to the ordering of variables in the identification scheme. Instead of the concept- based ordering implemented in the baseline specification, we choose a country-based ordering, where the U.S. are placed first. Within the U.S. part of the model, the variables follow the ordering as specified in Section 1.3.3, i.e. output and inflation are followed by the UMP variables, the VIX and the Fed balance sheet. All other variables of the model follow accordingly. This scheme features a stronger isolation of the U.S. from the rest of the world since U.S. monetary policy does not react to
any contemporaneous innovations abroad. The qualitative results are robust to this change in ordering. Another alternative ordering scheme that we apply focuses on the timing of portfolio flows. In particular, different to the baseline specification, we try an ordering scheme where all capital flow variables in the model are placed in front of the UMP block. This is done to allow for U.S. monetary policy to react instantaneously to portfolio outflows. On the other hand, this alternative ordering implies that the immediate response of portfolio flows to the UMP shock is zero. Despite the difference in the impact reaction, the overall response of U.S. portfolio outflows and EME portfolio inflows is still very similar to the baseline specification. This holds qualitatively and quantitatively.
Returning to our baseline ordering, we perform a number of additional robustness checks. For instance, we drop several countries one-by-one from the sample that display large reactions in one or more variables and find that mean responses change slightly quantitatively, but not qualitatively. We also perform the estimation using only a post-Lehman sample, i.e. starting in November 2008 instead of January 2008, and find no large qualitative changes resulting from this alteration. Likewise, in the baseline estimation we parsimoniously only include one lag in each country-specific VARX model. As a robustness exercise, we allow for a second or a third lag in the country-specific VARX models based on information criteria. Our findings are very robust to this exercise. A further robustness analysis is to include commodity price inflation as an exogenous (global) variable given that commodity price developments potentially play an important role for several EMEs. The results are qualitatively similar, but the inclusion of this exogenous variable produces smaller - but still significant - reactions of the the EMEs’ variables, except for portfolio inflows, which stay almost unchanged.
We also assess whether only one kind of portfolio assets, equity assets or bonds, drives the reaction of flows after a UMP shock. To do so, we replace the portfolio flow variables in our baseline model with equity flows and bond flows, respectively. We find that both equity and bond inflows into EMEs increase, with the rise in bond flows being larger on average. Also, from a cross-country comparison no obvious pattern arises which would indicate the relative importance of one type of flows over the other for single countries.
Further, to assess our identification strategy, we investigate how other U.S. vari- ables react to the UMP shocks in our model and compare the results to evidence from the literature. As found by Wright (2012) and Rogers et al. (2014), U.S. UMP causes a depreciation of the dollar and an increase in U.S. equity prices, among others. Hence, we add those variables separately to the U.S. part of the model and study the estimated response functions. Both responses look as expected given ex- isting evidence, but do not affect the responses of the baseline variables, and thus provide further support for our identification strategy.
As an alternative (unconventional) monetary policy instrument to identify the structural UMP shocks, we use the shadow federal funds rate constructed by Wu and Xia (2016). Specifically, we replace the UMP block, namely the VIX and the balance sheet, in our baseline model with the shadow rate and then apply a recursive ordering scheme to identify the UMP shock. Doing this yields qualitative similar responses for the EME variables as our baseline model with the identification scheme based on Gambacorta et al. (2014).
Lastly, we replace our preferred measures of financial conditions from the main specification one-by-one by other proxies that potentially similarly capture financial developments, and assess whether the reaction of the other variables is robust to this alterations. First, in the model FC, we replace the real interest rate, our proxy for domestic monetary policy, by the real lending rate. The responses of capital inflows, the real exchange rate and equity returns remain unaffected compared to the baseline FC model. The lending rate, on the other hand, displays a reaction very similar to the response of the interest rate in our baseline model, reflecting a close connection of the two rates in our sample. Second, following Bowman et al. (2015), we replace the real interest rate in the model FC by the long-term government bond yield. As before, we find that the reaction of flows, exchange rates, and equity prices is robust to this alteration. The mean responses show that long-term government bond yields significantly decrease, in line with findings by Bowman et al. (2015).
Third, we re-estimate the model FC replacing the interest rate by foreign exchange reserves growth. We do so as the accumulation of reserves is a policy tool actively used by central banks in EMEs, for instance to alleviate exchange rate appreciation pressures. Again we find that the reaction of portfolio flows, equity returns, and exchange rates is robust to this alteration. Moreover, reserve accumulation increases
in response to the UMP shock. Similar to the real interest rate response in the baseline model, this indicates that monetary policy in EMEs reacts to the U.S. expansion. In particular, the increase in reserves could be related to policies aimed at mitigating a currency appreciation that arises from the capital inflows.