key: cord-0675773-l0j1ixva
authors: Halouskov'a, Martina; Stavsek, Daniel; Finance, Mat'uvs Horv'ath Department of; Economics, The Faculty of; Administration,; University, Masaryk
title: The role of investor attention in global asset price variation during the invasion of Ukraine
date: 2022-05-12
journal: nan
DOI: nan
sha: bade8fc84f8aaa5a5eb3df975bd5fa8e2215b310
doc_id: 675773
cord_uid: l0j1ixva

We study the impact of event-specific attention indices - based on Google search queries - in predictive price variation models before and during the Russian invasion of Ukraine in February 2022. We extend our analyses to the importance of geographical proximity and economic openness to Russia within 51 global equity markets. Our results demonstrate that attention to the conflict is significant at the onset of and during the invasion and helps predict volatility. Finally, we find a positive dependency between attention significance and the geographical distance to Moscow and a negative dependency on the degree of economic openness to Russia.

The prelude to the Russo-Ukrainian war began in mid-October 2021, during which Russian forces gathered near Ukraine's borders and in the occupied Crimea region (Lister, 2021) . Shortly thereafter, aggressive and escalating statements from Russian policymakers were reported (Bowen, 2022) , which led Joe Biden to announce consequences in the event of any Russian invasion of Ukraine (Shalal et al., 2021) . As a result (on December 8), we observe the first sharp growth in the conflict attention index depicted in Figure 1 , meaning that this threat was likely recognized by many. The intelligence provided by Western security agencies suggested that a possible Russian invasion could start in early 2022 (Harris and Sonne, 2021) . Considering those warnings, combined with the growth of tensions, military movements (Salama et al., 2022) , and accusations (Olearchyk et al., 2022) , the attention paid to the possible conflict grew rapidly.

In his speech on February 21, 2022, President Putin announced that Russia had recognized the separatist republics in eastern Ukraine, which was followed by consequent threats to Ukraine (Reuters, 2022a) . In the early morning of February 24, Russian armed forces began a full-scale war against Ukraine.

The response was to impose unprecedented economic sanctions on the Russian economy, which were implemented shortly after the invasion and targeted all kinds of industries such as banking, oil exports, and high-tech components.

However, due to economic interconnectedness and dependency on Russian commodities, those sanctions also created risks for all companies and households in countries economically linked to Russia. The sanctions are already resulting in recognizable economic consequences, especially for Europe, whose largest energy supplier is Russia (in 2021, approximately 45% of imported natural gas came from Russia (IEA, 2022) ). Along with a sharp increase in energy prices, we are also witnessing devastating consequences for countries that rely on Ukrainian wheat imports (Simon and Davies, 2022) . Thus, this war has evolved into a serious global economic and political issue that is subject to exceptional worldwide attention. In this context, our aim is to investigate this extraordinary interest and determine whether it is linked to increased volatility of stock markets around the globe. We build upon a growing literature related to the relationship between future market movements and investor behavior, particularly investor sentiment and investor attention.

From a theoretical perspective, we rely on the limited attention hypothesis of Barber and Odean (2008) , according to which investors face a difficult task of choosing among numerous investment opportunities, despite possessing limited time and resources, and thus gravitate toward "attention-grabbing" options. Andrei and Hasler (2015) expand on this idea to demonstrate that news that receives substantial interest takes less time to be incorporated into prices. Their results also suggest that investors seek more information during times of high uncertainty -such as the unexpected Russian military invasion of Ukraine.

To capture investors' panic, fear, and uncertainty, we use a very popular direct measure of attention, the Google Search Volume Index (SVI). Some of the first applications of Google query data in research came from epidemiology (Ginsberg et al., 2009; Dugas et al., 2013) ; however, such approaches rapidly spread to the field of finance (Da et al., 2011; Joseph et al., 2011) as a proxy for attention (albeit sometimes imprecisely referred to as a sentiment proxy).

Search volume has been shown to be correlated with lagged trading volumes (Preis et al., 2010; Bordino et al., 2012) , to improve trading strategies (Preis et al., 2013; Bijl et al., 2016) or diversification strategies (Kristoufek, 2013) , and to be a driver of future market volatility (Aouadi et al., 2013; Vlastakis and Markellos, 2012; Hamid and Heiden, 2015; Dimpfl and Jank, 2016; Audrino et al., 2020) . Several studies have also examined interest in cryptocurrencies, measured with Google, as a driver of their price and volume fluctuations (Kristoufek, 2013 (Kristoufek, , 2015 Garcia et al., 2014; Cheah and Fry, 2015; Cretarola et al., 2017; Urquhart, 2018; Aalborg et al., 2019; Eom et al., 2019; Burggraf et al., 2020; Chen et al., 2020) . Similar results can be found for major FX markets (Kita and Wang, 2012; Smith, 2012; Goddard et al., 2015; Han et al., 2018; Wu et al., 2019; Saxena and Chakraborty, 2020; Kapounek et al., 2021) .

Investor attention appears to be particularly effective in studies related to specific events, which is also the case for our study, namely attention devoted to macroeconomic developments (Lyócsa et al., 2020b; Plíhal, 2021) , earnings news announcements (Hirshleifer et al., 2011; Hirshleifer and Sheng, 2021; Fricke et al., 2014; Ben-Rephael et al., 2017) , the outbreak of the COVID-19 pandemic (Chen et al., 2020; Lyócsa et al., 2020a) , and even this military conflict (Lyócsa and Plíhal, 2022) .

We contribute to this literature on the impact of event-specific attention by exploring the predictive power of investor attention devoted to the military conflict in Ukraine in volatility models. We divide the data into two samples (the pre-invasion and onset-of-invasion periods) to compare the effects of such attention on price fluctuations. 1 The analysis covers the stock indices of 51 countries. Our goal is to compare these results and determine whether (1) geographical or (2) economic proximity to the conflict influences the impact of conflict attention on volatility.

The remainder of this paper is organized as follows. In section 2, we describe the data, their sources, and how they were processed into the measures used for analysis. In the following section 3, we present and describe the results and emphasize their place in the context of the war and economic connectedness. Finally, we conclude by describing our results and contributions.

Our financial data consist of two datasets -the daily prices of selected indices and economic indicators. The first dataset includes data from all countries with an available MSCI index. MSCI indices were selected because of the consistent methodology used to calculate indices for all included countries. In addition to the MSCI indices, we include the Latvian stock index -OMX Riga -to capture the impact in this Baltic state. The data were collected from a Bloomberg terminal as an OHLC 2 dataset. After removing countries with missing OHL observations, our sample covers 51 countries, including 8 regions in the Americas, 3 regions in Africa, 19 European countries, and The second financial dataset considers a filtered set of countries based on previously mentioned conditions. For each country, we extracted data on imports from Russia and exports to Russia, as well as GDP, all for the year 2020. The source of these data is the UN Comtrade Database. This dataset was used to calculate the degree of openness (DOO) (Rodriguez, 2000) to Russia of country i country as follows:

where GDP i is the GDP of country i, Export i is the exports of country i to Russia and Import i is the imports of country i from Russia.

1 We use January, 1 2022, as the date to divide the data based on an undisclosed U.S. intelligence report warning of a Russian invasion of Ukraine in early 2022, first published in the Washington Post on December 3, 2021 (Harris and Sonne, 2021) . Thus our second sample captures the period of heightened attention and risk of an upcoming invasion. 2 An abbreviation for open, high, low, close data.

Our attention measures were retrieved from Google Trends with the help of the R package gtrendsR (Massicotte and Eddelbuettel, 2021) . Unfortunately, the availability of daily data is limited to 270-day intervals, and longer samples would require additional scaling. Although this may appear to be a relatively short sample, we opt for the 270-day interval, as it sufficiently covers the events we want to consider.

We use two sets of search terms to construct two variables, one related to general stock market attention and one for the attention paid to the military conflict, denoted G t and C t , respectively. We include the general index to ensure that we measure the effect of excessive attention to the conflict adjusted for the general day-to-day interest of investors in trading. Since we sought to capture global interest, we opted for queries of topics -an option that automatically translates keywords into all available languages and accounts for spelling variations. The G t and C t indices are then adjusted for the time zones of the stock exchanges corresponding to each MSCI index in our dataset.

For the MSCI indices that cover more than one country, we select the time zone with the majority coverage, as reported in the country weights in the MSCI fact sheets. Table 1 provides summary statistics for indices G t and C t after log transformation in three different time zones.

After accounting for time zones, we also remove values for nontrading days. Our approach consists of taking the maximum value of Friday to Sunday and assigning this value to Friday, and the method is applied for holidays.

This procedure must be applied individually to each country, as the nontrading days are not identical. Notes: S.D. stands for standard deviation, ρ(1) represents the first-order autocorrelation coefficient and ρ(5) denotes the fifth-order autocorrelation.

We used five topic search terms to create the conflict index C t ('Russia', 'Ukraine', 'Vladimir Putin', 'NATO', and 'sanctions') and 31 topics 3 to construct the general attention index G t . Both attention indices are then calculated as simple averages of the individual variables. These data take the form of a normalized volume ratio on a scale of 0-100, where 100 represents the maximum search activity during the selected period. The acquired search volume ratio SV I t at time t can be further transformed according to the procedure proposed by Da et al. (2011) into the abnormal search volume index (ASV I t ), which should help us to identify significant changes in Google searches.

The ASV I t is usually applied to address the noisiness of SV I t and to capture only abnormal search activity.

However, it does not capture the scale of the abnormal change. As the data we are working with have a rather unusual shape, with nearly exponential growth around the time of the invasion of Ukraine, we concluded that the ASV I t transformation is not suitable for our analysis. Instead, we opt for a log transformation of SV I t :

for the remainder of the paper we use G t and C t in this log-transformed form.

As we rely on MSCI indices, we are limited to the use of daily OHLC data, which do not allow us to use the standard realized volatility estimator calculated as the sum of squared intraday returns (see, e.g., Andersen et al. (2003)). Thus, to maintain the stylized facts 4 of the volatility time series, we employ a range-based volatility estimator following the approach of Lyócsa et al. (2021) , which was originally motivated by the approach of Patton and Sheppard (2009)q who notes that since the true data generating process is unknown, the optimal estimator must also be unknown. Therefore, a combination of several estimators may be less prone to estimator choice uncertainty:

where V t is the price variation estimator. Furthermore, for every day t = 1, 2, ..., T , we compute the average of three different realized range-based estimators (P K t , GK t , RS t ) and adjust the final price variation estimates for overnight price variation J t . We denote by P K t the estimator of Parkinson (1980) :

by GK t the volatility estimator of Garman and Klass (1980) is defined as:

by RS t the Rogers and Satchell (1991) estimator takes the form:

and finally, the overnight price variation is defined as:

, where O t , H t , L t , and C t represent the open, high, low and close prices on a given day t. Furthermore, define

The estimated daily volatilities are in A.6.

With a focus on the attention variables and subsequent exploration of their impact on price variation, we define a parsimonious model, mimicking the well-known and time-tested heterogeneous autoregressive (HAR-RV) model of Corsi (2009) . In contrast to the standard HAR-RV, we omit the monthly component because, as presented in Table A .6, the fifth-order autocorrelation is low in some cases. During our main period of interest, the uncertainty about subsequent price developments is so high that what the last month's price variation was should rarely matter.

where C t and G t are the attention variables defined in the previous section. V w t is the weekly price variation component given as V w t = 5 −1 4 j=0 V t−j . The model in Equation 8 is estimated via ordinary least squares and for both data samples, that is, in the pre-invasion and onset-of-invasion periods. Next, we estimate the model with the log-transformed price variation.

First, we estimated the model defined in Equation 8 for all 51 stock market indices for both the before the invasion and onset-of-invasion periods. The results show that in 70% of countries, the conflict attention variable C t has a significant positive effect on future volatility. In other words, the more common military conflict topic searches are, as measured by Google searches, the higher the next day's volatility of MSCI indices. Table 2 then presents estimates and diagnostics for those indices for which we found the most significant impact of conflict attention, while Table 3 shows the least significant impacts. The parameter estimates of the remaining countries are reported in the Appendix in Table A.4 and Table A .5. Each table also compares the results for the sample before the invasion of Ukraine (Panel B) and during the onset of the invasion (Panel A).

In some cases the model diagnostics show mild heteroskedasticity and autocorrelation of residuals as indicated by the p-values of the White and Ljung-Box tests (see Tables 2, 3, A.4 and A.5). To overcome these issues, we applied the Newey-Further-West estimator (Newey and West, 1987) . We decided to apply this method to all indices because this facilitates comparing the resulting t-statistics across estimated models. We tested the explanatory variables for the presence of a unit root via the test of Pesaran (2007) for panel data stationarity. This test rejected the null hypothesis of a unit root across the cross-section of the volatility series. Some of the estimated models possess a low R 2 value, which may be a result of a low persistence of the estimated volatility proxy in the pre-invasion sample.

In models reported in Tables 2 and A.4, the conflict attention measures essentially replace the traditional role of daily volatility, which is apparent in the substantially higher R 2 values than those observed in the before invasion period. Figure 2 summarizes the results of our study. The dots mark the countries analyzed, with their color determined by the p-value of conflict attention variable C t and their size determined by the magnitude of these parameter estimates. We can see that conflict attention primarily affects the volatility in European countries, where the conflict increases volatility. The countries outside of Europe are rarely significantly affected and show a lower impact on their indices' volatility.

Based on the results in Table 2 , the day-ahead volatility is most significantly affected by conflict attention at the onset of the invasion period in ten European countries. In addition, we find that all European countries in Notes: On the y-axis, we use the absolute value of the t-statistic for estimated parameter Ct during the onset-of-invasion sample. We excluded Latvia from the chart for better graphical representation because its DOO value places it far from the other observations. However, this decision does not bias the presented results in any way.

our dataset except for Norway are highly impacted. In fact, the effect is concentrated in Europe and near the military conflict, which is also visible in Figure 2 . We assume that these countries might be influenced due to strong economic ties with Russia, the threat of wider European conflict with the possible involvement of NATO or the ongoing humanitarian crisis (with a few million refugees fleeing Ukraine). Of the countries facing a wave of refugees, our analysis covers Poland, Czechia, Hungary, and Germany, all of which are among the most impacted countries. The previously mentioned dependence of Europe on oil and gas imports from Russia may also be an important reason why this conflict has primarily affected European stock markets. This argument is particularly supported by the fact that we have not found a significant impact on Norway, which is independent of Russia due to the former's substantial oil and gas reserves.

However, in the pre-invasion period, the conflict attention variable was not significant at the 5 % level for any of the countries considered. This statement is also valid for wartime data for the countries extracted in Table 3 . These are primarily American and Asian countries. Regarding the variable representing the previous day's volatility V t , despite the significance of this parameter for many countries in the before invasion period, V t became insignificant for most of the indices during the invasion.

There are also some significantly impacted countries outside of Europe, for instance, Mexico, whose results could be explained by the fact that Mexico did not condemn the Russian invasion (Reuters, 2022b) . Various reasons could explain the impact of the conflict in other countries. We have, for example, countries that are dependent on wheat imports from Ukraine, or those, that like Russia, extract oil.

Based on these findings, we assume that the effect of the conflict attention variable increases the geographically closer the country is to Russia or the stronger its relations are with Russia. We decided to graphically verify this relationship in Figure 3 , which displays the relationship between the conflict attention t-value of each country and its economic openness with Russia and between the conflict attention t-value and its geographical distance from

Moscow.

Based on these charts, we conclude that the more open a country is to Russia, meaning a higher ratio of Russian imports and exports to the country's GDP, the more significant that conflict attention is. Conversely, conflict attention is less significant for countries that are more geographically distant from the Russian capital.

Achieved results and their significance may be influenced by several factors of choice we had to make in order .

where we assign D t with ones for the first week of invasion starting on 21st February and also for 21st and 24th

February separately. The conflict variable remained significant.

This paper explores how investors' attention to the conflict between Russia and Ukraine influences the variability of asset prices in specific countries. We construct a Google search-based military conflict attention variable and general stock market attention to capture the effect of excessive attention devoted to the conflict. To draw sharp conclusions about volatility, we applied an HAR-RV model with a range-volatility estimator to MSCI data.

Our results demonstrate that while the impact of the conflict attention measure was insignificant in the preinvasion period, at a time of escalating war threats, attention to conflict significantly affects volatility. Specifically, increasing conflict attention leads to higher volatility of the indices of the studied countries. The analysis of the indicators of the economic and geographical interconnectedness of individual countries to Russia shows that the effect of attention is more significant in countries with higher openness with Russia and those nearer to it. Notes: In the header, we use ISO 3166 country codes. Regression parameter estimates in bold indicate significance at the 10% level; superscripts a, b, c and d denote statistical significance of estimated coefficients at the 10%, 5%, 1% and 0.1% level. Residuals ρ(1) describe the first-order autocorrelation of the residuals. For the Ljung-Box and White tests, we report the corresponding p-values. Notes: In the header, we use ISO 3166 country codes. Regression parameter estimates in bold indicate significance at the 10% level; superscripts a, b, c and d denote statistical significance of estimated coefficients at the 10%, 5%, 1% and 0.1% level. Residuals ρ(1) describe the first-order autocorrelation of the residuals. For the Ljung-Box and White tests, we report the corresponding p-values.

What can explain the price, volatility and trading volume of bitcoin?

Modeling and forecasting realized volatility

Investor attention and stock market volatility

Investor attention and stock market activity: Evidence from france

The impact of sentiment and attention measures on stock market volatility

All that glitters: The effect of attention and news on the buying behavior of individual and institutional investors

It Depends on Where You Search: Institutional Investor Attention and Underreaction to News. The Review of Financial Studies

Google searches and stock returns

Web search queries can predict stock market volumes

Russian Military Buildup Along the Ukrainian Border

Do fears drive bitcoin? Review of Behavioral Finance

Speculative bubbles in bitcoin markets? an empirical investigation into the fundamental value of bitcoin

Fear sentiment, uncertainty, and bitcoin price dynamics: The case of covid-19. Emerging Markets Finance and Trade

A simple approximate long-memory model of realized volatility

A sentiment-based model for the bitcoin: theory, estimation and option pricing

In search of attention

Can internet search queries help to predict stock market volatility?

Influenza forecasting with google flu trends

Bitcoin and investor sentiment: Statistical characteristics and predictability

Google search, information uncertainty, and post-earnings announcement drift

The digital traces of bubbles: feedback cycles between socio-economic signals in the bitcoin economy

On the estimation of security price volatilities from historical data

Detecting influenza epidemics using search engine query data

Investor attention and fx market volatility

Forecasting volatility with empirical similarity and google trends

Does investor attention matter? the attention-return relationships in fx markets

se-2?accountid=16531. name -North Atlantic Treaty Organization-NATO; Copyright -Copyright WP Company LLC d/b/a The Washington Post

Limited Investor Attention and Stock Market Misreactions to Accounting Information

Macro news and micro news: Complements or substitutes

International energy agency: How europe can cut natural gas imports from russia significantly within a year

Forecasting abnormal stock returns and trading volume using investor sentiment: Evidence from online search

Selective attention in exchange rate forecasting

Investor attention and fx market volatility

Can Google Trends search queries contribute to risk diversification?

What are the main drivers of the bitcoin price? evidence from wavelet coherence analysis

Satellite photos raise concerns of russian military build-up near ukraine

North Atlantic Treaty Organization-NATO; Copyright -Copyright 2021 Cable News Network

Yolo trading: Riding with the herd during the gamestop episode

Fear of the coronavirus and the stock markets

Impact of macroeconomic news, regulation and hacking exchange markets on the volatility of bitcoin

Russia's Ruble during the onset of the Russo-Ukrainian war: The role of Implied volatility and Attention

2021. gtrendsR: Perform and Display Google Trends Queries

A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix

Us accuses russia of planning 'false-flag operation' in eastern ukraine

United States-US

The extreme value method for estimating the variance of the rate of return

Optimal combinations of realised volatility estimators

A simple panel unit root test in the presence of cross-section dependence

Scheduled macroeconomic news announcements and Forex volatility forecasting

Quantifying trading behavior in financial markets using google trends

Complex dynamics of our economic life on different scales: insights from search engine query data

Extracts from putin's speech on ukraine

Mexico declines to impose economic sanctions on russia

On the degree of openness of an open economy

Estimating variance from high, low and closing prices

Russia says it sees little scope for optimism in u.s. proposals on ukraine -wsj

Should we pay attention to investor attention in forex futures market?

Biden warns putin of sanctions, aid for ukraine military if russia invades

we need bread': fears in middle east as ukraine war hits wheat imports

Google internet search activity and volatility prediction in the market for foreign currency

What causes the attention of bitcoin?

Information demand and stock market volatility

Our currency, your attention: Contagion spillovers of investor attention on currency returns