COVID-19 Stringency Measures, Risk of Openness and Foreign Investment: Some Preliminary Evidence

This paper investigates the relationship between foreign investment and the containment measures, in the aftermath of the COVID-19 spread. We find that the non significant correlation between the average quarterly stringency index and inward foreign investment at the end of the first quarter of 2020 hides a source of heterogeneity across countries. Foreign investors highly rate the implementation of strong containment measures - as measured by government stringency index - in countries with high risk - as measured by the risk of openness index. Conversely, foreign investors are less attracted by assets issued by countries adopting weak stringency measures despite a high risk of openness, or those implementing drastic stringency measures in the presence of a relatively lower risk of openness.


Motivation and relevance
As the COVID disease has spread around the world, many government have been forced to impose restriction policies with di¤erent intensity and timing ): some countries have rapidly introduced very strict measures in the immediate aftermath of the crisis, such as total lockdown, and then have removed them, as a consequence of an improvement in the transmission; other countries instead reacted with more gradual and punctual interventions, as soon as small outbreaks occurred.
The economic and …nancial consequences of these restrictions cannot be properly evaluated yet, and any attempt of making predictions can be incautious: the literature on the impact of epidemics on …nancial markets is indeed scarce, and all parallels with other natural disasters or terroristic attacks are bound to be unreliable (Godell (2020)). However, it is evident that the COVID outbreak has already lead recession, erosion of con…dence and higher uncertainty (OECD (2020b)).
The growing recent literature about the impact of the COVID outbreak on …nancial markets generally converges on the evidence of a signi…cant impact of COVID con…rmed cases or deaths on …nancial markets'volatility and liquidity (Albulescu (2020); Baig et al. (2020); Salisu and Vo (2020); Ashraf (2020)).
As far as international investments are concerned, Saurav et al. (2020) highlight that the COVID-19 crisis represents for international enterprises a new and unprecedented source of investor risk that is depressing investor con…dence. OECD (2020a) and OECD (2020c) assess that foreign direct investments of …rms are expected to decline sharply as a consequence of the pandemic and of the stringent public health measures to limit the spread of the COVID-19, with a notable heterogeneity across countries. Emerging economies have already experienced a massive drop of portfolio investment in ‡ows, because international investors transfer capital back home, or invest in safer assets during periods of uncertainty. Giofre'(2020a) con…rms that, within a generalized decline in foreign investment, advanced countries, with higher GDP per capita, members of the G7 group, or of the Euro area are signi…cantly less severely hit by the pandemic than emerging and developing countries. Kizys et al. (2021) study the e¤ects of the Oxford COVID-19 Government Response Tracker, whose higher scores are associated with greater stringency, on herding behavior in international stock mar-kets during the coronavirus COVID-19 outbreak. They disclose the presence of herding behavior in the …rst three months of 2020 in 72 countries stock markets' countries, but also highlight that this herding behavior is mitigated by more stringent government response to the coronavirus crisis, by way of reducing multidimensional uncertainty. Giofre'(2020b) …nds that the average stringency index is not correlated with inward investment positions, but the within-country standard deviation of the stringency index is positively and signi…cantly correlated with inward portfolio investments at the end of the …rst quarter of 2020: foreign portfolio investments, typically more volatile and reactive than foreign direct ones, are more responsive to governments'prompt reactions than to gradual ones at the end of the …rst quarter, thus suggesting that the former policies might be perceived as a more serious commitment to stem the spread of COVID-19.
This paper contributes to the literature by further investigating the presence of a signi…cant cross-country heterogeneity in the relationship between foreign investment and the government containment measures, at the end of the …rst quarter of 2020. Investors scrutinize sources of risk and the means through which this risk can be reduced. Foreign investors could be therefore deterred from or attracted by investing in a country adopting more radical stringency measures, depending on how these policies are perceived in terms of commitment to stability, lower uncertainty, and then higher adjusted risk-returns in the near future. We …nd indeed that the non signi…cant correlation between inward foreign investment at the end of the …rst quarter of 2020 and the average quarterly stringency index hides an important heterogeneity across countries: the response of investors'behavior to the adoption of severe measures depends indeed on the level of pandemic risk in the economy.
Speci…cally, the extent to which severe containment measures -as measured by government stringency measures (SI) -signi…cantly a¤ect the inward foreign investment in a country would crucially depends on the country level of risk directly connected with the non adoption or removal of these stringency measures -as proxied by the risk of openness index (ROI).
The empirical evidence actually shows that foreign investors, at the end of the …rst quarter of 2020, value the assets issued by countries, which calibrate the stringency measures to the risk of openness. On the one hand, they highly rate the implementation of strong containment measures in countries with high risk of openness. On the other hand, they appear to avoid investing in those countries adopting weak stringency measures despite a high risk of openness, or those implementing drastic stringency measures in the presence of a low risk of openness.
The remainder of the paper is structured as follows. In Section 2, we sketch the estimable equation. In Section 3, we describe the data, and provide some descriptive statistics. In Section 4, we report the results of the empirical analysis. Section 5 concludes.

Estimable equation
Our objective is to establish the evolution of international investments in the immediate aftermath of the adoption of COVID-19 containment measures.
We empirically test the existence of a relationship between the average stringency index in a country and its end-of-period (quarter/semester) foreign liabilities.
Let's de…ne, …rst, the growth of liabilities ( L) in the …rst quarter (q1) of 2020 as q1: it is the di¤erence between the liabilities at the end of the …rst quarter (March 2020, L 03_20 ) and the We compute this growth in liabilities also for the …rst semester ( s1) of 2020.
where L 06_20 is the value of foreign liabilities at the end of the second quarter of 2020.
In our main speci…cation, we regress the growth in foreign liabilities on the average within-country stringency index (SI); risk of openness index (ROI), and their interaction (SI ROI), as follows: We are mainly interested in the sign and size of the , and coe¢ cients. If the adoption of severe stringency of containment measures (SI), or a high risk of openness (ROI); deter foreign inward investment, then we should observe a signi…cant negative ; or ; coe¢ cient, and vice versa.
If the risk of openness (ROI) in one country a¤ects the way severe stringency measures (SI) are perceived and valued by foreign investors, then we should observe a signi…cant coe¢ cient.
We trade-o¤ a parsimonious speci…cation, due to the low number of observations, with the need to include time-varying regressors, which might contribute to explain the growth in foreign investments, and covariates potentially correlated with our main regressors, whose exclusion could bias the estimated coe¢ cients. It is worth stressing that, we can ignore any country-speci…c …xed e¤ects, as these are removed by the construction of the dependent variable in di¤erence form.
We include, …rst, the (lagged) appreciation in the Nominal E¤ective Exchange Rate (NEER), because its change might have a¤ected foreign investment. Second, we control for the number of new COVID-deaths and its within-country standard deviation, as the stringency index is potentially strongly correlated with the health indicators of the epidemic. Finally, we include two binary indicators of economic and …nancial development, to control, for instance, for the presence of any eventual ‡ight to quality propensity by foreign investors.
We consider two alternative de…nitions of the dependent variable in the analysis. Beyond the measure of liabilities'growth derived in equation (1), we add the measure dif f L, in the attempt to address the issue of seasonality of foreign investment allocations: it is derived as the di¤erence between the 2020 L measure, as de…ned in equation (1), and the corresponding measure in 2019.
For instance, dif f q1 is the …rst quarter measure, and is de…ned as follows: and, analogously, dif f s1 is de…ned for the semi-annual period as: To estimate the parameters in equation (3), we adopt, in the baseline speci…cation, a Robust Least Squares estimation. Ordinary least squares estimators are sensitive to the presence of observations that lie outside the norm for the regression model of interest. The sensitivity of conventional regression methods to these outlier observations can result in coe¢ cient estimates that do not accurately re ‡ect the underlying statistical relationship. Robust least squares refers to a variety of regression methods speci…cally designed to be robust, or less sensitive, to outliers. Among Robust least squares, we adopt the M-estimation developed by Huber (1973). 1 Alternative estimation methods, such as standard OLS and Quantile regressions, are run for comparison with our baseline results.

Data and descriptive statistics
We consider foreign inward investment (or foreign liabilities) in 53 countries, upon data availability.
Data are drawn from the International Investment Position Statistics, released by the IMF, which provides information on foreign assets and liabilities, classi…ed in several categories and instruments, at a quarterly frequency. For most of the analysis, we consider Foreign Total Liabilities (FTL), but in the last table we also consider its sub-components, Foreign Direct and Foreign Portfolio Liabilities.
The source of COVID-related data is a Github ongoing repository of data on coronavirus, the Coronavirus Open Citations Dataset.
We draw from this dataset our main regressors relying on the Oxford's Coronavirus Government Response Tracker (OxCGRT) ): the stringency index (SI), which represents a proxy for the severity of the containment policy measures adopted and the Risk of Openness Index (ROI), based on the recommendations set out by the World Health Organisation (WHO) of the measures that should be put in place before Covid-19 response policies can be safely relaxed. We also include data about new COVID-deaths and cases per million of inhabitants. All these data are originally reported at a daily frequency, but we construct the corresponding quarterly averages, in order to match the quarterly frequency of the dependent variable.
We include in our speci…cation other three controls. First, the NEER (Nominal e¤ective exchange rate, broad index), released by the Bank for International Settlements. Then, we include two binary indicators of economic and …nancial development, i.e., the GDP per capita and the market capitalization per GDP, drawn from CEIC data.
In Figure 1 and 2, we report the distribution an main descriptive statistics of the dependent variable. Figure 1 considers the growth of FTL in the …rst quarter of 2020, while Figure 2 considers the growth of FTL in the …rst semester of 2020. Panel a) in both …gures relies on the measure de…ned in equation (1) and (2), while panel b) relies on the dif f measure de…ned in equation (4) and (5). We can observe, …rst, that the measure is more negatively skewed in the …rst quarter, than in the …rst semester. Second, the distribution of the measure dif f , de…ned in equation (4) and (5), is more negatively skewed than the measure, in both the …rst quarter and the …rst semester.
In Figure 3, we report the distribution and main descriptive statistics of the average quarterly Stringency Index (SI) in the …rst and second quarter.
The average stringency index, whose original values range 0-100, is about 19 in the …rst quarter, while in the second quarter grows to 71, thus pointing to a sharp tightening of the anti-COVID 19 containment measures in the second quarter of 2020. Figure 4 reports the distribution and main descriptive statistics of the average quarterly Risk of Openness (ROI) in the …rst and second quarter. While the mean slightly increases in the second quarter of 2020, the median in almost una¤ected, and the standard deviation decreases.
In Table 1 In Table 2, we report the main …ndings of our multivariate regression analysis for the …rst quarter, under a Robust Least Squares estimation. The dependent variable is the growth in Foreign Total Liabilities (FTL), as de…ned in equation (1), at the end of quarter (or semester). As anticipated in Section 2, the fact that the dependent variable is de…ned in di¤erence form, allows us to ignore any problem related to country-speci…c …xed e¤ects, removed by construction. We are forced to keep a parsimonious speci…cation, because we can rely on a quite limited country sample. We include, on the one hand, time-varying regressors which might concur to explain the growth in foreign liabilities, and, on the other hand, covariates potentially correlated with our main regressors, whose exclusion could bias the estimated coe¢ cients.
Our main regressor is the Stringency Index (SI), based on the Oxford's Coronavirus Government Response Tracker, and is computed as the quarterly average of daily data. We include, as a …rst control, the (one-month lagged) growth in the Nominal E¤ective Exchange Rate (NEER), a measure of the appreciation of the economy's currency against a broad basket of currencies, because its change might have a¤ected foreign investment in the country. 3 Second, we control for the quarterly average of new COVID-deaths per million of inhabitants: as shown in Table 1, the stringency index is correlated with this speci…c indicator of the epidemic, as it represents the government reaction to contain new cases, deaths, and intensive-care treatments. 4 Finally, we include two binary indicators of economic and …nancial development, GDP per capita 2 New COVID-cases and new COVID-deaths are always included as alternative controls. To account for the sig-ni…cant correlation of the SI and ROI indexes with the new COVID-death per million (or new COVID-cases per million), we have also performed our regression analysis excluding these covariates, with no signi…cant e¤ect on our results. 3 We include its one-month lagged value, to avoid endogeneity issues. 4 Our …ndings are left qualitatively unchanged, when considering a regression speci…cation with the new COVIDdeaths per million of inhabitants in logs, rather than in levels. and market capitalization to GDP. Since country speci…c factors are swept away by construction of the dependent variable in di¤erence form, we do not include the level of development of individual countries. We consider, instead, two indicators of development, equal to 1 if the country belongs to the developed group, and 0 otherwise, in order to control for the presence of a ‡ow of foreign investments towards high versus low developed countries: according to the ‡ight to quality rationale, for instance, in the presence of a global shock, foreign investors would deviate their investments to more stable and developed economies. 5 We can notice that in column (1) of Table 2 the coe¢ cient of the stringency index is not signi…cant, and the only signi…cant coe¢ cients are the ones related to economic and …nancial development.
In particular, the coe¢ cient of the economic development indicator is positive and statistically significant, while the coe¢ cient of the …nancial development dummy variable is negative and (marginally) signi…cant. These latter …ndings are in line with the results in Giofre' (2020a), which reveal that investment in less developed countries have been more dramatically reduced by the COVID outbreak. It is worth noticing that while the positive e¤ect of the economic development dummy (high_GDP_cap) is more reliable, since stable and signi…cant across di¤erent regression speci…cations, the negative coe¢ cient of the MCAP/GDP binary variable, as already pointed out in Giofre' (2020a), is instead not reliable, since unstable in terms of signi…cance and sign across regression speci…cations.
In column (2) of Table 2, we add the quarterly average Risk of Openness Index (ROI) as a covariate, in order to check whether the risk of relaxing containment measures and opening to social and economic activities has a¤ected the growth in foreign liabilities. We observe that both indexes are not signi…cant determinants of the growth in total foreign liabilities in the …rst quarter.
We conjecture however that the non signi…cant e¤ect of the stringency measures on foreign liabilities could hide a cross-country heterogeneity. We test in particular if the response of one country's inward foreign investment to its government'containment policy depends upon the risk of openness faced by the same country. Investors are interested in monitoring sources of risk and the means to reduce it, and their behavior can be particularly reactive to government actions aimed at challenging severe risks. We conjecture therefore that the importance of strong containment measures -as measured by government stringency measures (SI) -could crucially depend on the level of risk directly connected with the non adoption or removal of these measures -as measured by the risk of openness index (ROI).
We add to our regression model the interaction between the stringency index SI and the risk of openness index ROI: if our hypothesis is correct, we should observe a positive signi…cant coe¢ cient of the interaction term, thus suggesting that foreign investors consider the impact of the containment measures more e¤ective in economies with a higher risk of openness.
While the results in column (3) do not fully support our initial conjecture, they point to a promising direction. The coe¢ cient of ROI is negative and signi…cant (-0.383), while the coe¢ cients of SI and of the interaction term SI ROI are not statistically signi…cant. 6 However, the non signi…cant coe¢ cients are close to statistical signi…cance, and their sign is consistent with our predictions: the coe¢ cient of SI is negative with a p-value equal to 0.121, while the coe¢ cient of SI ROI is positive and its p-value is 0.104.
In column (4), we replace the number of "new COVID-deaths per million of inhabitants" with the covariate "new COVID-cases per mn of inhabitants". Ashraf (2020) …nds that stock markets reacted more proactively to the growth in number of con…rmed cases as compared to the growth in number of COVID deaths. We …nd that our …ndings are qualitatively una¤ected by the introduction of this alternative covariate, in terms of signi…cance and size of coe¢ cients.
The explanatory note of the ROI index speci…es that ‹ ‹the OxCGRT data cannot say precisely the risk faced by each country, it does provide for a rough comparison across nations. Even this "high level"view reveals that many countries are still facing considerable risks as they ease the stringency of policies ››. We check whether accounting for the natural "measurement" error implicit in the construction of any index, and then considering it less "literally", helps supporting our hypothesis.
We construct therefore a binary variable for each of the two indexes, ROI and the SI, splitting countries into those above and those below their respective mean.
Interestingly, we observe in column (5a) that the results corroborate our hypothesis, when both indexes are dichotomized. To interpret the e¤ects, it is worth considering that the default-excluded group in the regression is the subset of countries with a below the mean SI and ROI. We observe …rst that, in the …rst quarter of 2020, countries with a high SI and a low ROI feature a 5.79% lower growth in FTL, while countries with high ROI and low SI display a 4.3% lower growth in FTL. Those countries adopting high SI in the presence of high ROI display instead a signi…cantly larger growth in inward foreign investment, as shown by the coe¢ cient of the interacted term (10.54%). To seize the overall growth in FTL for countries with high SI and high ROI relative to countries with high SI and low ROI, we need to add the coe¢ cient of the interaction term to the "pure" e¤ect of the SI index: compared to countries with high SI and low ROI which witness a decrease in FTL by 5.79%, those countries with high SI and high ROI display a higher FTL by 4.75% (=-5.79%+10.54%).
This …nding supports our conjecture that foreign investors are in ‡uenced by the implementation of stringency indexes as long as these measures are meant to reduce a high risk of openness. Conversely, foreign investors in general appear to discard the assets issued by those countries whose containment policies are relatively mismatched with the risk of openness, that is, those adopting weak stringency measures despite a high risk of openness, or those implementing drastic stringency measures in the presence of a lower than average risk of openness.
In column (5b), we check whether our …ndings are sensitive to a di¤erent speci…cation of the threshold -median instead of mean-for de…ning the binary version of the two indexes, and we observe that the …ndings are substantially unchanged.
Finally, columns (6a) and (6b) replicate the regressions of columns (5a) and (5b), when the "number of new COVID-deaths per million" is replaced by the "number of new COVID-cases per million", and results are con…rmed, with only modest changes in the size of the coe¢ cients. Table 3 replicates Table 2, but the dependent variable relies on the dif f measure, whose structure is de…ned in equation (4), rather than on the measure. This measure aims to address the issue of the seasonality of foreign investment allocations, as it is derived as the di¤erence between the 2020 measure, and the corresponding measure in 2019.
This table provides results very similar to the ones reported in Table 2. In the …rst two columns, we observe no signi…cant coe¢ cients, but in the third column, we …nd some hints in support of our hypothesis, as the coe¢ cients of the SI and of the ROI index are negative, while the coe¢ cient of the interaction term SI ROI is positive. However, when replacing the "new COVID-deaths" control with the "new COVID-cases", all coe¢ cients, with the exception of ROI's, fall below the standard bar of statistical signi…cance. As in Table 2, only when considering the indexes in a dichotomic version, the sign and signi…cance of coe¢ cients get consistent with our conjecture. In column (5a), we observe that in the …rst quarter of 2020, countries with high SI and low ROI witness a decrease in FTL by 6.41% with respect to 2019, while those countries with high SI and high ROI indeed display an increase in FTL by 2.09% (=-6.41%+8.50%).
A comparison with Table 2 reveals that the e¤ect is still present and statistically signi…cant, thus supporting our hypothesis, although the e¤ect on the growth of FTL in the …rst quarter of 2020, as a di¤erence to the corresponding quarter of 2019, is halved in size. Results are con…rmed, when considering the median threshold rather than mean for the construction of the binary indexes of SI and ROI (column (5b)), or when considering the alternative covariate "new COVID-cases" under both speci…cation of the threshold (columns (6a) and (6b)). Table 4 reports the results of Table 2 and 3, but relative to the end of the second quarter, so that the dependent variable becomes the growth in liabilities at a one-semester distance. For the sake of brevity, we report only the relevant regressors. The upper part of the table (panel I) refers to the measure, while the bottom part (panel II) refers to the dif f measure, de…ned as the di¤erence between the growth of FTL in the …rst semester 2020 with respect to the …rst semester in 2019. We observe that, di¤erently from the …rst quarter, the SI and ROI indexes have no signi…cant role in driving foreign investors'decisions at the end of the second quarter, under any speci…cation of the indexes, or of the dependent variable. 7 7 We only observe in panel I, some sparse and non systematic negative signi…cant coe¢ cients of the ROI index.
The empirical evidence shows that the measures of containment and the risk of openness may help explain the foreign investors'choice in the immediate aftermath of the COVID outbreak, but not in the second quarter. 8

Sensitivity checks and additional analysis
In the remainder of the paper, we undergo our …ndings to a bunch of robustness checks and additional studies, to understand the strengths and limits of the analysis.
In Table 5, we check the sensitivity of our …rst quarter signi…cant …ndings to the estimation strategy and speci…cation of the country sample. Table 4, the upper part of the The …rst three columns, consider alternative estimation models, while columns (4) to (5c) consider di¤erent country sample speci…cations. At the head of the rows, we specify that the indexes are de…ned in dichotomic form (SI_d and ROI_d). In column (1) of panel I and II, we report, for comparability, the results of column (5a) of Table 2 and Table 3, following the Robust Least Squares baseline approach.

As in
Column (2) reports the results of the regression under an OLS speci…cation, which are qualitatively similar to the ones in column (1). Columns (3) report the results of a Quantile regression.
The quantile regression estimates are more robust against outliers in the response measurements: whereas the method of least squares estimates the conditional mean of the response variable, quantile regression estimates its conditional median (or other quantiles). We show that, in the …rst panel, the results relative to the median of the response variable are fully in line with our previous …ndings, both in terms of signi…cance and size. Conversely, panel II, where the dependent variable is de…ned as the di¤erence between the growth of FTL in the …rst quarter of 2020 and the corresponding growth in 2019, displays less robust results, with a (marginally) signi…cant coe¢ cient of the interaction term SI ROI; and a (marginally) non signi…cant coe¢ cient of the SI term (p-value 0.11).
In columns (4) to (5c) of panel I and II, we test whether our …ndings survive to the exclusion of speci…c countries from the sample.
In column (4), we exclude China. China has been the …rst country to be struck by the COVID spread, several weeks before other countries. The estimate of the stringency measures on foreign investors' choices at the end of the …rst quarter could therefore have been signi…cantly driven by China's asynchronic timing of lockdown and loosening measures.
By comparison with column (1), we observe that the exclusion of China, in both panels, hardly a¤ects the size and signi…cance of the e¤ects, which remain still sizeable and signi…cant.
In columns (5a) to (5c) of Table 5, we exclude from the sample potential o¤shore …nancial centres, to make sure our results are not driven by economies distorting investors'decisions for reasons hard to control in our analysis. We consider three di¤erent classi…cations proposed by the literature: column (5a) reports the results under the classi…cation proposed by Damgaard et al. (2018), column (5b) follows Zoromé (2007), while column (5c) follows Lane and Milesi-Ferretti (2017) (see Appendix A.1, for details on the countries excluded). By comparison with column (1), we observe that qualitatively the results are con…rmed, and interestingly, the exclusion of o¤shore centres even reinforce them: both the negative coe¢ cient of the SI index and the positive coe¢ cient of the interaction term SI ROI are larger in size, and even more statistically signi…cant.
In the two following tables, Table 6 and 7, we perform additional analysis, to match our …ndings with the evidence of the recent empirical literature on the evolution of foreign investment, after the spread of the COVID contagion.
In particular, OECD (2020a) and OECD (2020c) assess that foreign direct investments of …rms are expected to decline sharply as a consequence of the pandemic, with a notable heterogeneity across countries, and that emerging market economies have indeed already experienced a massive drop of portfolio investment in ‡ows, because international investors transfer capital back home, or invest in safer assets during periods of uncertainty. Giofre'(2020a) con…rms that, within a generalized decline in foreign investment, advanced countries, with higher GDP per capita, members of the G7 group, or of the Euro area have been signi…cantly less severely hit by the pandemic than emerging and developing countries.
In Table 6, we check whether accounting for di¤erent country grouping invalidate or modify our …ndings. In columns (1a) and (1b), we report, for comparison, the results of columns (5a) and (5b) of Table 2. Columns (#a) report results when the two OxCGRT indexes are recoded as binary indicators with the mean as threshold, while in columns (#b) the threshold is the median. We consider di¤erent alternative to the GDP per capita to de…ne the group of advanced economies following Giofre'(2020a), and we also …nd that foreign investment in these countries have been less severely hit by the crisis, as the coe¢ cients of the dummy associated with advanced, G7 and Euro area countries are positive and statistically signi…cant, also after accounting for the binary OxCGRT indexes and their interaction.
We observe that when we include as control a dummy for advanced economies, results are very close to our baseline; when we include, as a control, the G7 group, the coe¢ cients of the SI and SI ROI are signi…cant only is dichotomized around the mean, though the size of the interaction term's coe¢ cient is smaller; …nally, when controlling for the Euro area country group, we observe consistent coe¢ cients of the interaction terms (though smaller in size), but no signi…cant negative coe¢ cient of the SI regressor (i.e., for countries with high SI and low ROI): Finally, we check how our results are a¤ected by the inclusion of the standard deviation of the stringency index ( SI) in our regression speci…cation. Giofre'(2020b) …nds that the within-country standard deviation of the stringency index is positively and signi…cantly correlated with foreign inward investments, at the end of the …rst quarter of 2020, but only as far as portfolio investments are concerned. She suggests that foreign portfolio investors, typically more reactive than foreign direct investors, could have be more responsive to governments'prompt reactions than to gradual ones in the immediate aftermath of the outbreak, since the former policies could have been perceived as a more serious commitment to stem the spread of COVID-19.
Our model speci…cation provides an alternative explanation to the non signi…cant e¤ect of the SI index on foreign investment: we check within a uni…ed framework if the two pieces of empirical evidence are consistent or mutually exclusive. If countries with a higher standard deviation in SI are also those with a higher risk of openness, by omitting one of the two factors makes the coe¢ cients of the included covariates biased. 9 We consider a regression speci…cation including both regressors, to shed light on the drivers of the growth in foreign inward investments. Since Giofre' (2020b) underlines a di¤erent role for portfolio and direct inward investments (that is, foreign direct and foreign portfolio liabilities), we consider their respective growth, together with the growth of total foreign inward investments.
In Figure 5, . We observe that the growth for FPL is more negatively skewed than the distribution of FDL.
To allow an immediate comparison across di¤erent types of liabilities, we report the coe¢ cients of our regressors of interest in Table 7, which is horizontally partitioned into three panels: panel I refers to FTL, panel II to FDL, and panel III to FPL.
Let's focus, …rst, on how our …ndings are a¤ected by the inclusion of the SI as a regressor, and, then, on how the results in Giofre'(2020b) are altered by imposing our framework.
In panel I, the regression setting is the same as in Table 2, with the addition of the standard deviation of the stringency index ( SI) as a regressor. We observe that the coe¢ cient of the SI regressor in the …rst two columns is not signi…cant, as in our previous …ndings. Also after including the interaction term between SI and ROI, we do not observe any signi…cant coe¢ cient of SI, either when controlling for the new COVID-deaths (column (3)) or when controlling for the new COVIDcases (column (4)). When we instead recode the SI and ROI indexes in a dichotomic form (columns (5a) to (6b)), we …nd again a statistically signi…cant coe¢ cient of the interaction term SI ROI; consistently with our hypothesis, while the negative coe¢ cient of the SI is statistically di¤erent from zero only when the threshold of the dummy is the mean (columns (#a)). 10 9 The correlation of the …rst quarter SI with the average stringency index SI is equal to 0.18 (and only marginally signi…cant: p-value=0.098), while its correlation with the average risk of openness index ROI is equal -0.03 (but not statistically signi…cant at any conventional level). 10 A coe¢ cient of the SI regressor not statistically di¤erent from zero, implies that countries with a high SI and The evidence on the SI regressor is fully in line with the results of Giofre' (2020b) for FTL, both in terms of sign (positive), signi…cance (marginal), and size (about 0.3%) of the associated coe¢ cient.
In panel II and III, we compare Foreign Direct Liabilities and Foreign Portfolio Liabilities. Giofre' (2020b) …nds that foreign direct inward investment have shown a lower responsiveness to SI, than foreign portfolio inward investments. Indeed, also in our setting, we con…rm that the coe¢ cient of the SI is almost twice as large for FPL than for FDL, and its statistical signi…cance is much stronger and systematic.
Let's now consider how the interaction of the two OxCGRT indexes may have di¤erently a¤ected the growth in foreign direct and foreign portfolio liabilities.
When considering the dichotomic version of the SI and ROI indexes in columns (5a) to (6b), we observe the following di¤erences between FDL and FPL.
On the one hand, the coe¢ cient of the interaction term SI ROI is signi…cant only for columns (#a) for FDL, while is always statistically di¤erent from zero for FPL (though only marginally, when the threshold is computed over the median); on the other hand, the overall e¤ect of countries with high SI and high ROI is larger for foreign portfolio liabilities than for foreign direct liabilities. Indeed, though the negative coe¢ cients of the SI regressor and the positive coe¢ cients of the interaction term SI ROI are larger (in absolute value) for FDL than for FPL (when both are signi…cant), the overall net e¤ect of high SI in economies with high ROI on foreign liabilities' growth is smaller for FDL than for FPL: in column (5a)  Moreover, we observe that in columns (3) and (4)  so far, the growth in foreign portfolio liabilities in the …rst quarter of 2020, is signi…cantly associated also with the continuous version of the stringency index SI, of the risk of openness ROI, and of their interaction. It means that, while the growth in foreign direct liabilities only respond to high versus low indexes, foreign portfolio liabilities are tilted by a cross-country marginal di¤erence of the stringency index within economies with a varying degree of openness risk exposure. For instance, the results in column (4) of panel III, can be read as follows: an increase in the SI index, originally ranging from 0 to 100, by 1 unit leads to a lower growth in FPL by 1.46% if the risk of openness ROI is set at its minimum, that is equal to 0, while the same increase in the SI index by 1 unit leads to an increase in the growth of FPL by 0.7% (0.0070=-0.0146+0.0217), when the level of risk of openness is set at its maximum, that is equal to 1. This evidence points to a tighter and closer responsiveness of foreign portfolio liabilities to the stringency measures adopted and to the COVID risk exposure of the country, and is in line with the results in Giofre'(2020b), con…rming a general higher reactivity by foreign portfolio investors than by direct investors in the immediate aftermath of the crisis.
Overall, the results of Table 7 show that our …ndings and the evidence in Giofre'(2020b) are both consistent within a more general framework: on the one hand, the role of the standard deviation of the stringency index SI survives as a signi…cant driver of foreign portfolio investment, and, on the other hand, the inclusion of a potentially correlated regressor, such as SI, does not a¤ect the validity of our …ndings, and even enriches the analysis, by unfolding a peculiar sensitivity of foreign portfolio investors to the adoption of COVID containment measures at the end of the …rst quarter of 2020.

Conclusions
This paper investigates the evolution of foreign investment in the immediate aftermath of the adoption of stringency measures to restrain the spread of COVID-19. Investors scrutinize sources of risk and the means through which this risk can be reduced. Foreign investors could therefore be deterred from or attracted by investing in a country adopting more radical stringency measures, depending on how these policies are perceived in terms of commitment to stability, lower uncertainty, and then higher adjusted risk-returns in the near future. We …nd that the non signi…cant correlation between the average quarterly stringency index and inward foreign investment at the end of the …rst quarter of 2020 hides a source of heterogeneity across countries: the response of investors'behavior to the adoption of COVID restrictive policies depends indeed on the level of pandemic risk of the economy.
Speci…cally, the extent to which severe containment measures, as measured by the government stringency index, signi…cantly a¤ects the inward foreign investment in a country depends on its level of risk connected with the non-adoption or removal of these stringency measures, as proxied by the risk of openness index.
The empirical evidence relative to the end of the …rst quarter of 2020 actually shows that foreign investors value the assets issued by countries, which calibrate the stringency measures to the risk of openness. On the one hand, they highly rate the implementation of strong containment measures in countries with high risk of openness. On the other hand, they appear to avoid investing in those countries adopting weak stringency measures despite a high risk of openness, or those implementing drastic stringency measures in the presence of a low risk of openness.
It is worth emphasizing that the objective of this research is to establish the existence of a connection between COVID restrictive measures and foreign investors' allocation choices, and not to question the appropriateness of the containment measures imposed by di¤erent countries: the policies adopted represent the country-speci…c reactions to country-speci…c conditions, in terms of severity of cases, deaths, and pre-existing e¢ ciency of the national health system. Indeed, severity and speed of adoption of policies in di¤erent countries can be strictly related to the severity of the e¤ects of the COVID spread, which has shown a remarkable cross-country heterogeneity, for reasons that are worth investigating in the near future. This paper emphasizes the importance of taking into account this multifaceted heterogeneity, by considering how the diversity in the risk of openness across countries has a¤ected the relationship between government containment policies and foreign investment decisions.

Tables Figure 1. Quarterly growth in Foreign Total Liabilities
This …gure reports the distribution and main statistics of the growth in Foreign Total Liabilities at the end of the …rst quarter 2020. Figure 1a) refers to the growth measure q1, as de…ned in equation (1), while …gure 1b) refers to the growth measure dif f q1, as de…ned in equation (4). To enhance readability, growth rates are reported in percentage.

Figure 2. Semi-annual growth in Foreign Total Liabilities
This …gure replicates Figure 1, but is referred to the …rst semester of 2020, rather than to the …rst quarter. To enhance readability, growth rates are reported in percentage.

Table 1. Correlation matrix of COVID regressors
This table reports the correlation matrix of COVID-related regressors. The upper panel refers to the …rst quarter of 2020, while the second one refers to the second quarter. Statistically signi…cant Pearsoncorrelation coe¢ cients are reported in bold characters (t-test p-values in square brackets).

Table 2. Main …ndings: q1
This table reports the results of a Robust Least Squares regression (M-estimation), following equation (3). The dependent variable is the quarterly growth in Foreign Total Liabilities, at the end of the …rst quarter of 2020, de…ned as in equation (1). In columns (5a) and (5b), the Stringency Index (SI) and the Risk of Openness Index (ROI) are re-decoded as binary variables 0-1, if, respectively, their average quarterly value is, respectively, below or above the mean (column (5a)), or median (column (5b)). ***, **, and * indicate signi…cance at the 1, 5, and 10% levels, respectively.

Main findings diff∆q1
SI & ROI (0-1) > mean > median > mean > median   This table replicates Table 2 and 3, but relative to the …rst semester of 2020. In panel I, the dependent variable is constructed following the structure of equation (1), while in panel II, the dependent variable follows the structure of equation (4). The econometric speci…cation is the same as in previous tables: it also includes the controls reported at the bottom of the   (1), while in Panel II the dependent variable is dif f q1, as de…ned in equation (4). For the sake of brevity, only results with binary de…nition of SI_d and ROI_d (1 if above the mean, 0 otherwise) are reported. Column 1 of Panel I and II reports, for comparability, the results under a Robust Least Squares estimation of column (5a) of Table 2 and 3, respectively. Column 2 and 3 report the results of the same regression under a standard OLS estimation and a Quantile regression computed at the median of the dependent variable. Columns (4) to (5c) report results when excluding China (columns (4)), or o¤shore countries (from columns (5a) to (5c)) from the sample. O¤shore countries are de…ned according to three alternative o¤shore de…nitions: column (5a) follows the classi…cation in Damgaard et al. (2018), column (5b) follows Zoromé (2007), columns (5c) follows Lane and Milesi-Ferretti (2017)

Figure 5. Quarterly growth in Foreign Direct and Portfolio Liabilities
This …gure reports the distribution and main statistics of the growth in Direct and Portfolio Liabilities at the end of the …rst quarter 2020, as de…ned in equation (1). To enhance readability, growth rates are reported in percentage. In our analysis, we consider and report as regressors both the quarterly overall mean of the daily stringency index (SI) and its quarterly standard deviation ( SI), in Table 7, computed within each country over the corresponding quarter.
Source: https://github.com/OxCGRT/covid-policy-tracker Risk of Openness Index (ROI) The Oxford COVID-19 Government Response Tracker (OxCGRT) provides a cross-national overview of the risk and response of di¤erent countries as they tighten and relax physical distancing measures. The Risk of Openness Index is based on the recommendations set out by the World Health Organisation (WHO) of the measures that should be put in place before Covid-19 response policies can be safely relaxed. Considering that many countries have already started to lift measures, the Risk of Openness Index is a reviewed version of our previous 'Lockdown rollback checklist'.
While the OxCGRT data cannot say precisely the risk faced by each country, it does provide for a rough comparison across nations. Even this "high level"view reveals that many countries are still facing considerable risks as they ease the stringency of policies.

Other controls
New COVID death per mn This is a daily variable, reported by the countries'authorities. In our analysis, we consider both the quarterly average of new COVID-19 deaths and its standard deviation, computed within each country over the corresponding quarter.
Source: https://github.com New COVID cases per mn This is a daily variable, reported by the countries'authorities. In our analysis, we consider both the average quarterly number of new COVID-19 cases and its standard deviation, computed within each country over the corresponding quarter.
Source: https://github.com Nominal E¤ective Exchange Rate BIS e¤ective exchange rate Nominal, Broad Indices Monthly averages; 2010=100. The NEER regressor is included with the same structure as the dependent variable. For instance, if the dependent variable is q1 as de…ned in equation (1), then the regressor included is (N EER 03_20 N EER 12_19 )=N EER 12_19 Source: Bank for International Settlements High MCAP/GDP Market capitalization to GDP (year: 2019, or latest available data). The regressor included is a binary variable equal to 1 if the market capitalization per GDP is larger than the sample mean, and 0 otherwise. Source: CEIC data High GDP per capita The regressor included is a binary variable equal to 1 if the GDP per capita is larger than the sample median, and 0 otherwise. GDP per capita (year: 2019, or latest available data). Source: CEIC data Advanced, G7, Euro area