Hawks and Doves: Financial Market Perception of Western Support for Ukraine

........................................................................................................................ v


Abstract
Since the Russian invasion of Ukraine on February 24, 2022, the West has been intensively discussing its support strategy. Hawkish positions of strengthening Ukraine via armaments, financial resources, and sanctions against Russia compete with dovish views of avoiding further escalation of the military and geopolitical conflict. News from the war became a dominating factor for international politics and the world economy. In this paper, we analyse how international financial markets perceived this news, especially on Western positions. We create a comprehensive data set of news related to the war and measure reactions of five key financial markets. The results show that stronger support for Ukraine had a positive impact after the first weeks of the war when the Ukrainian position in the war improved, but a negative or at least less positive influence before. Thus, financial markets seem to have perceived support as a risk of further escalation threatening global economic activity in the first phase. However, a hawkish line was a positive signal for financial markets after the change in perceptions. The results also confirm that the war and escalation in general had harmful effects on international financial markets.

JEL-Classification: G14, H56
Keywords: Conflict, Event Study, Financial Markets, News, Russia, Sanctions, Ukraine We are grateful to Taras Shypka and Yasemin Yilmaz for valuable support and to Artem Dyachenko for comments on an earlier version of the paper. The dataset on war-related news is available online at: https://www.uni-trier.de/fileadmin/fb4/prof/VWL/EWF/Data_Sets/Dataset_UA_War_News.xlsx.

Introduction
Since February 24, 2022, the Russian war against Ukraine has dominated the headlines and the international political agenda. From the local nature, which was characteristic for battles through 2014-2022, the events very soon unfolded into a full war. Many Western governments support Ukraine with armaments, financial resources (Antezza et al. 2022), and sanctions against Russia without a direct involvement in the battlefield. Beyond the tragic local consequences, war, sanctions, and the energy crisis are having an immense impact on the global economy (e.g. Garicano et al. 2022). This concerns many sectors such as energy, agriculture, and commodities. For example, energy-intensive manufacturing was immediately affected after the start of the war , prices showed a quick and strong reaction (Ozili 2022), and trade flows were redirected ).
News of war events have become important not only locally but also for global financial markets. Empirical evidence shows that conflicts in the world typically affect stock and commodity prices (Schneider andTroeger 2006, Guidolin andLa Ferrara 2010). Before the Russian attack in 2022, the Russian-Ukrainian war has been shown to affect both Russian and Ukrainian stock markets in 2014 (Hoffmann and Neuenkirch 2017). The 2022 Russian war against Ukraine was not an exception. Researchers unveiled the effect of geopolitical risk on market prices depending on those markets' distances from the conflict region (Federle et al. 2022, Chiţu et al. 2022, Hossain and Masum 2022. With regard to the Russian attack, appropriate reactions have been intensely discussed in the West. Concerning financial aid, military aid, and sanctions, typically hawkish and dovish positions have to be distinguished. Our work sheds light on the significance or war-related news for financial markets, especially focusing on such decisions of the West. We uncover in how far specific news are conceived as positive or negative signals from the viewpoint of the world economy. Military aid, financial aid, and sanctions may be seen either as stabilizing the Ukrainian position or as steps of further escalation threatening global economic activity. We cover five key financial markets, namely US and European stocks (mirroring a general business outlook), oil and gas for the energy dimension, and wheat representing agriculture.
In this paper, we create a systematic data set on news from Ukraine linked to the Russia-Ukraine war. Our sample period spans from February 24 to July 12, 2022. Besides military aid, financial aid and sanctions, news categories comprise combat escalation, the export situation (concerning agriculture, mining, and logistics), and the local situation (concerning resilience, humanitarian issues, and economic development). The latter categories are crucial for our analysis as control variables. For instance, it is likely that news on sanctions often coincide with escalation, so escalation must be controlled for in order to isolate the effect of sanctions. Our comprehensive news data set allows for such differentiations.
Regarding military aid, financial aid, and sanctions, it is of special interest if perceptions changed over time as the war saw dramatic turns and geopolitical beliefs were re-evaluated. For instance, was adopting a hawkish line seen more critically in the first phase of the war and more positively when Ukraine recorded successes? Accordingly, we check for the potential change in perception of the different types of the news over time in the five financial markets by performing automated breakpoint tests. 1 Indeed, we find that market perception varied over time and that aid and sanctions are received much more positively in the second subsample. That is, clear support for Ukraine is now a positive sign for international markets, while it was less so in the first weeks when war prospects were worse and support may have been seen as an escalation risk. In a similar vein, Chiţu et al. (2022) show that for the first 14 days into the war, the effect of the physical proximity to Kyiv was more visible in stock returns than 40 days into the war. Importantly, our results also confirm that the war and escalation in general had harmful effects on international financial markets. Finally, we document the robustness of our results with respect to an alternative timing of the war-related news and potential non-linearities in the markets' response to these. We also show that it is news about sanctions that matters for markets and not the number of actually sanctioned Russian entities. The same holds for news about combat escalation as compared to actually observed territorial gains or losses.
1 A related branch of the existing literature examines time-variation in asset price responses. A popular focus hereby is to distinguish the reaction during expansions and recessions (e.g. Andersen et al. 2007) or with a varying degree of volatility in financial markets (e.g. Ehrmann and Fratzscher 2005). Other time patterns, such as a secular decline or a structural break, are also considered in the literature (e.g. Faust et al. 2007, Fratzscher 2009, Ehrmann et al. 2011). An early contribution using a time-varying parameter approach is provided by Cocco and Fischer (1989) and a more general test for time variation is given in Goldberg and Grisse (2013). However, our sample period is too short for these (more sophisticated) methods to detect time variation in the response of financial markets. With less than 100 observations and several categories of war-related news we have to resort to automated structural break tests.
The remainder of this paper is structured as follows. Section 2 outlines the course of the war and potential breakpoints in detail. Section 3 describes the creation of the news data set and the empirical methodology. Section 4 discusses the estimation results. Section 5 concludes.

The course of the war
Within our sample period, the course of the war has made an important turn from the fast-paced advance of the Russian forces during the first month to the successful defensive operations of Ukrainian forces and (even) counteroffensive operations later on. Our hypothesis is that the markets' perception of the war also has changed. Accordingly, the reaction on war-related news should be different after a (yet to be determined) breakpoint. A more detailed description and mapping of the course of the war can be found in Annex 2. Additional background information on Western aid and sanctions as well as the economic situation in Ukraine can be found in Annex 3.
The first weeks from February 24 to early April were marked by Russian rapid advance.
Within less than a month, Russian troops occupied parts of the Sumy, Chernihiv, and Kyiv The slow but steady shift in the global perception of the war and Ukraine's ability to fight back might be best illustrated by the increasing complexity and firepower of weapons that the West was ready to provide to Ukraine. This is also underscored by Figure 1 that illustrates the search interest of Ukrainian users for certain types of military assistance over time.
A similar shift in the global perception occurred in the economic and political sphere. After several months of fiscal struggles in Ukraine and domestic monetary expansion to cover the gap in the state budget, Western financial aid became more regular and predictable and helped to stabilize the state finances (cf. Antezza et al., 2022). Another important turn for Ukraine wasafter some initial scepticism -the acceptance as European Union (EU) candidate country in June.

Figure 1: Search interests of Ukrainian users for certain types of military assistance
Notes: Index is set to 100 for the peak popularity. Source: Vyshlinsky et al. (2022), updated for 2023 data.
Regarding economic sanctions on Russia, the US, the EU, and other countries were also moving gradually. The EU has been Russia's largest economic partner and relied heavily on Russian energy exports (in particular natural gas), whereas the US never had a substantial economic relationship with Russia in the first place (Congressional Research Service 2022).
This results in lower (expected) costs of sanctions for the US and might explain that the US launched oil and gas embargo procedures in Senate in early March. 4 In contrast, the EU has been hesitant and only included a coal embargo into the 5th package of sanctions in early April and a phased-out oil embargo (for maritime supplies) into the 6th package of sanctions in early June.
To summarize, the course of the war changed over time and Western economic and political support for Ukraine (including sanctions on Russia) steadily increased after some initial hesitance. Therefore, one could also expect a shift in financial markets' perception of war-related news. To check for such a breakpoint in perception during our sample period February-July 2022, one should -based on the considerations above -focus on the period April-May. The automated statistical check for breakpoints (see below in Section 3) confirms this prior.

Data on war-related news
In a first step, we build a country-level dataset for Ukraine containing news from February 24 to July 12, 2022. The news are collected from the Interfax news agency. 5 We chose Interfax (economic section) as it is a primary non-governmental source with direct access to a broad range of information related to Ukraine. It is one of the three oldest news agencies in Ukraine, working since 1992. Interfax is beneficiary-owned by its management with no known relations to Russia and is being used as a primary source of information by Ukrainian investment banks and journalists writing about Ukraine in international media. We also consider it a good source for Western aid and sanctions, since these decisions from various countries directly affect Ukraine and are receiving a wide coverage. However, regarding this data, we will also conduct robustness checks with other sources. A caveat of this or any other data source might be that it is not covering all the relevant events in Ukraine. However, given its reputation and detailed coverage of economic and political events, we rely on the assumption that it covers a significant majority of them. Importantly, news agency data, compared to online analytical or economic/political online media sources, minimises the risk of fake news and covering a large number of irrelevant articles.
The English-language Interfax Ukraine economic section yields a total of 7,109 news items over that period. We classify the data manually using a two-staged process. First, we assign a "sectoral code" to each news item and, second, we assess the "direction" of the event for each news story. After classifying each news story separately, we do a double-checking process and review each set of news items with a separate sectoral code and direction to ensure these indeed belong to the same group.
Our approach to classifying the news items includes a breakdown along the following lines (see Table 1 below). It allows us to separate military, local, and international types of news (1st code symbol) and then smaller segments within the broader categories (2nd code symbol). We also assign each news item a direction (binary, 3rd code symbol  We select all news items that belong to these categories and sorted out unrelated items (e.g. local public events announcements, occasional news from unrelated foreign countries, data on market quotations of bonds, interest rates etc.) and the items without a clear direction or with very minor, virtually absent implications (outdated news items based on pre-war press releases, politicians' encouragements and biddings, and irrelevant or minor news). The final news items set contains 4,371 news items. After the initial categorization, we merge the finer categories into larger news items clusters, which yields the classification in Table   1. The more detailed representation of the categories can be found in Annex 1, Tables A1.3 and A1.4.
Plausibly, we find a relatively low number of news on less military aid and less financial aid.
The low number does not invalidate these variables, but of course, sample variation increases for them. Still, we will find statistically significant effects, but the high variation has to be taken into account when interpreting sometimes comparatively high coefficient values.

Empirical methodology and financial market data
With regard to news on the war, we analyse the reactions of stock and futures markets.
These expectations are, in turn, determined by the available information set at time t ( ): Similarly, the (logarithmic) price of a future (log ) is a conditional expectation (ℎ) of the price of an asset (log in period t+1 given the available information set at time t ( ): Accordingly, changes of the stock price in Eq. (1) and the expected asset price in Eq.
(2), that is, the returns, are a function of a change in the information set. Such changes in the information set are driven by news on, for instance, global economic conditions. Any warrelated news might influence the local and global environment and would also affect the returns of stocks and futures ( ).
Consequently, we employ the news indicators as key explanatory variables in a set of linear regressions in a second step. The general specification is as follows: ( 3) is the return of a financial series of interest, a constant term, the lagged return to control for market persistence, the change in the US three-month T-bill to proxy for changes in the global monetary policy environment (following, e.g. Kutan 2005, Fender et al. 2012), the vector of war-related news described in Section 3.1, and is the error term. Eq. (3) is estimated with least squares and heteroskedasticity-and autocorrelation robust standard errors. 6 As financial series, we utilize the returns of the S&P 500 and the Eurostoxx 50 for a US (or global) and European stock market perspective. Returns of ICE Brent Futures and Dutch TTF Natural Gas 1 Month Gas Futures cover the main energy markets, whereas returns of Chicago SRW Wheat Futures are included to account for the agricultural sector. 7 In particular, the choice of wheat futures is motivated by the role of Ukraine as major supplier of these goods, whereas both energy series are accounted for because of Russia's role in their supply.
All series are close-to-close returns and trading stops at 4:00 pm Eastern Time for all markets except the Eurostoxx, which closes at 6:00 pm Central European Time. This corresponds to a 7:00 pm Kyiv time closure of the Eurostoxx 50 and an 11:00 pm Kyiv time closure of all other markets. Almost all news items on Interfax are published before 7:00 pm Kyiv time. Hence, the news enters Eq. (3) on the day of its publication. News items that were published on a non-trading day were moved to the next trading day.
As we conjecture a potential change in the perception of war-related news by financial markets over time, we estimate Eq. (3) for the full sample period (February 24 to July 12) and test for a structural break. In our automated breakpoint detection procedure, we restrict the subsamples to a minimum of 30 observations, which yields a total of 35 potential breakpoints.
We then estimate two separate models for the subsamples that generate the lowest joint 6 Our sample is too short for a successful GARCH modelling approach. Indeed, GARCH(1,1) estimations do only converge for a part of the market considered in this paper. In addition, our breakpoint analysis is unfeasible with such an approach given the limited number of observations at hand. 7 Data is obtained from Bloomberg or the Wall Street Journal. For the very few missing values in the Eurostoxx 50 series, we used linear interpolation. Controlling for these imputed values with the help of dummy variables does not have a qualitatively relevant effect on our results. residual sum of squares (RSS). Figure A4.1 in Annex 4 plots the RSS with the endpoint of the first subsample on the x-axis. In the case of the Eurostoxx 50 (April 7), oil (April 6), gas (April 6), and wheat (April 8), the first subsample ends right after the liberation of Northern Ukraine. For the S&P 500 (16 May), we detect the breakpoint at a later time, that is, after the start of the Ukrainian counteroffensive north and east of Kharkiv. This rather late break date for S&P 500 compared to the other markets appears to be driven by days with multiple news in the same category. When we use a concave function of the news variable (i.e. a log plus one transformation; see Section 4.2 and Table A4.5), the break date results as early as April 12. In all cases, the RSS decreases by at least 24 percent when allowing for different coefficients across subsamples. 8 Hence, this procedure reveals changes in financial markets' response to war-related news. Table A4.1 in Annex 4 shows descriptive statistics for the five financial market return series (also split into two subsamples based on the test procedure outlined above). On average, we observe negative returns for both stock market series with no noticeable differences across the subsamples. In the case of oil and wheat futures, however, the returns are -on averagepositive in the first subsample and negative in the second period. This is indicative of the steep increase in the prices of these goods at the beginning of the war and the later observed normalization tendency. In the case of gas futures, Table A4.1 shows an increase in both subsamples, reflecting the tight conditions in this market ever since the start of the Russian invasion (which relaxed only later in the year). Turning to the volatility of the series, we find a lower standard deviation in the second subsample for the Eurostoxx, oil futures, gas futures, and wheat futures, which, however, is not resembled in the S&P 500. These patterns are confirmed by the time series plots in Figure A4.2, which further underline the need to analyse potential differences in the financial markets' perception of the war over time. However, it remains to be seen if the descriptive differences in the financial market return series can be explained by a change in the response to the war-related news over time.
Finally, we would like to emphasize that our analysis is based on a couple of assumptions.
First, we assume that there are no missing third-party factors correlated with the explanatory news variables. This assumption can be justified as the war in the Ukraine is the overarching topic in the first half of 2022 in international politics and financial markets. Developments, such as the rapid changes in in energy markets, were driven by the war events. In addition, we control for changes in the global monetary policy environment. Second, it is typically the unexpected component of an event that should actually matter for financial market participants. In an ideal world, we would extract such a news component from market expectations as it is done in the literature on macroeconomic news or monetary policy decisions (following, e.g., Fratzscher 2004, 2005) and their impact on financial markets. However, in the absence of such a series, we consider the actually observed news as second-best proxy. Nevertheless, it has to be mentioned that the number of news items for a given topic on a given day might serve as a good indicator for the intensity of news in that category.

Baseline results
Tables 2-6 show the results of Eq. (3) for both stock return series as well as the returns of oil, gas, and wheat futures. All left-hand side variables are measured in percent. Hence, the coefficients represent the effects of an additional item in a news category in percentage points.
We first focus on the variables related to the research question, that is, those reflecting Western support for Ukraine (military aid, financial aid, and sanctions).
News on more military aid leads to a decrease in S&P returns in the full sample period and the first subsample (columns 1 and 2 of Table 2) and an increase in in wheat futures in the first subsample (column 1 of Table 6). The latter effect corroborates the discussions of a potential shortage in the worldwide wheat supply, particularly during the early phase of the war. In a similar vein, news on less military aid leads to higher Eurostoxx returns (column 2 of Table 3) in the first subsample and lower US returns in the second subsample (column 3 of Table 3).
Opposing signs for the first and second subsample can also be observed for gas futures where less military aid for Ukraine was appreciated by participants in the first subsample, but lead to higher returns in the second one (columns 2 and 3 of Table 5). Lastly, wheat futures decreased after that type of news when considering the full sample period and -to a larger extent -in the first subsample (columns 1 and 2 of Table 6). The same pattern, even though not statistically significant, emerges for more military aid in the gas market and less military aid in the oil market.
News on more financial aid is found to decrease US stock returns in the first subsample (column 2 in Table 2) and European stock returns when considering the full sample period (column 1 in Table 3). The latter effect appears to be driven by the first subsample, where the coefficient is less precisely estimated, though. Still, these results resemble the response to military aid news as financial markets prefer less support in the first subsample and more support in the second subsample. However, for the remaining markets, estimation uncertainty for the financial aid effects is large. These less distinct results as compared to those for military aid indicate a clear pecking order in terms of the markets' response to different aid categories. (3) with heteroskedasticity and autocorrelation-robust standard errors in parentheses. ***/**/* indicate significance at the 1%/5%/10% level, respectively. Figure A4.1 shows the results for the breakpoint tests. (3) with heteroskedasticity and autocorrelation-robust standard errors in parentheses. ***/**/* indicate significance at the 1%/5%/10% level, respectively. Figure A4.1 shows the results for the breakpoint tests. (3) with heteroskedasticity and autocorrelation-robust standard errors in parentheses. ***/**/* indicate significance at the 1%/5%/10% level, respectively. Figure A4.1 shows the results for the breakpoint tests. (3) with heteroskedasticity and autocorrelationrobust standard errors in parentheses. ***/**/* indicate significance at the 1%/5%/10% level, respectively. Figure  A4.1 shows the results for the breakpoint tests. (3) with heteroskedasticity and autocorrelationrobust standard errors in parentheses. ***/**/* indicate significance at the 1%/5%/10% level, respectively. Figure  A4.1 shows the results for the breakpoint tests.
News on stronger sanctions generally leads to higher stock returns (columns 1-3 in Table 2 and Table 3). This mirrors a hawkish line, since overall business evaluation improves despite the fact the sanctions regularly imply also restrictions for domestic firms. News on weaker sanctions, which might be seen as a relief for the strained energy market, is reflected in lower gas returns in both subsamples (columns 2 and 3 of Table 5). This type of news is also associated with higher wheat returns (column 1 of Table 6 In a second step, we look into the response to other war-related news, such as combat (de-)escalation including the humanitarian situation, export-related news, and the domestic economic and political situation. Controlling for these variables is important for a clear answer to our research question since there might be correlation patterns (e.g. between conflict escalation and sanctions) that could confound our results. Indeed, an estimation of Eq. (3) without these categories (not shown, but available on request) yields a (much) worse fit as the one reported in Tables 2-6 and sometimes also (substantial) differences in the estimates for news on military aid, financial aid, and sanctions. 9 In general, news on combat escalation is bad news for stocks markets as indicated by the decline in US and European returns (columns 2 and 3 in Table 2 and columns 1 and 2 in Table 3) and for oil futures as reflected in the rise of returns during the second subsample (column 3 in Table 4). For the latter market, we also observe a decrease in returns after news on combat deescalation (columns 1 and 2 in Table 4). A similar easing of the strained conditions can be found for wheat future returns (column 1 in Table 6). In the case of news on humanitarian aid, we only find a significant negative response of US stock returns in the second subsample (column 3 of Table 2). Hence, conflict escalations (including a larger need for humanitarian aid) is bad for financial markets, whereas de-escalation is appreciated.
Turning to exports, we find a decrease in the returns of gas futures in the first subsample and the full sample period (columns 1 and 3 of Table 5) after positive news. In addition, we detect an increase in the returns of oil futures and gas futures in the first subsample (column 2 of Table   4 and Table 5) and wheat futures when considering the full sample period (column 1 of Table   6) after negative news. Accordingly, commodity markets are positively (negatively) affected if there are fewer (more) restrictions to the worldwide supply of goods.
Lastly, we look into the response of financial markets to other domestic economic and political news. Local news pointing towards a more resilient Ukraine leads to lower European stock returns and an increase in gas futures in the first subsample (column 2 of Table 3 and   Table 5). In a similar vein, we observe higher stock returns in the US (column 1 in Table 2) as well as lower returns of oil and gas futures in the first subsample (column 2 in Table 4 and Table 5) after local news indicating a more fragile Ukraine. These results fit the financial markets' response to Western news: signs of a resilient Ukraine were not appreciated during the first phase of the war.
The most important take away from the results of the controls, however, is that our key findings hold when accounting for a broad set of war-related news. We can conclude that financial markets had a more dovish stance during the first phase of the war and a more hawkish later on, while conflict and escalation in general had harmful effects on international financial markets over the full sample period. Lastly, our results are also not confounded by other exportrelated news or domestic economic and political news.
Thus far, we did not put much emphasis on the interpretation of the absolute size of the news effects. As an illustration, one might resort to the response of the S&P 500 to news about stronger sanctions. During the full sample period, an additional item in that category leads to an increase of 0.14 percentage points (pp). The corresponding effects in the first and second subsamples are 0.14 pp and 0.40 pp, respectively. These figures can be put into context by looking into the frequency of news items in Table 1 and the standard deviation of the financial series in Table A4.1. This underscores that the response to war-related news does not only feature a structural break over time. In addition, the estimated effects are of economic relevance.

Robustness checks
Next, we explore the robustness of our results along different dimensions. First, we test if our timing of war-related news is appropriate and allow all news items to enter Eq. (3) contemporaneously and additionally with a lag of one day. Given the large number of categories, this procedure is only feasible for the full-sample estimations. Table A4.2 compares the adjusted R2 and the Bayesian Information Criterion (BIC) for the models in Tables 2-6 and   models where we additionally include the lagged news variables. All augmented models are worse in these two statistics and an exclusion restriction on all lagged terms cannot be rejected at the 10 percent level. The only exception is the Eurostoxx 50 where the adjusted R2 of the augmented model is higher and the exclusion restriction is rejected (while the BIC is still worse when compared to the more parsimonious specification). The results of the larger model (available on request) indicate that the coefficients in column 1 in Table 3 are virtually unchanged by the additional regressors. In addition, the first lag of more (less) financial aid is found to be significantly negative (positive). Accordingly, these results fit in the general pattern of the full-sample market response to news (cf. Section 4.1).
Second, we test if our results remain robust once we control for the change in Ukrainian territory occupied by Russian forces. For this purpose, we create an indicator for occupied territory based on the maps provided by @War_Mapper on Twitter (cf. Annex 2) and include its daily change as additional covariate into Eq. (3). 10 The results can be found in Table A4.3.
They leave the point estimates of the news indicators for conflict (de-)escalation and their standard errors virtually unchanged. In addition, the indicator for occupied territory is significant only in the first subsample of the Eurostoxx 50 and the gas futures estimations. The positive coefficient on the stock market and the negative effect on gas futures are consistent with the general picture that financial markets perceived a quick Russian advancement not as a bad signal early in the war. To summarize, the results are reassuring in that our news variables retain the major effect even when controlling for the actual course of the war.
Third, in a similar exercise, we explore if our results for the response to news on sanctions hold when additionally controlling for actual sanctions that came into effect (cf. Figure A3.1).  Table A4.5 are very similar to those in Table 3, again with the response to sanctions in the second subsample being the only minor exception. In addition, the structural break in the Eurostoxx 600 (April 11) can be found at almost the same day as for the Eurostoxx 50 (April 7). Hence, the result patterns for European stocks can be found in a narrow and a broad stock index.
Finally, we consider potential non-linearities in the response of financial markets to news. It might be the case that markets respond differently to a change from, say, zero items to one in a given variable as opposed to a change from five to six. We impose such a concave pattern by 10 Colour filters were applied to each daily map and adjusted according to their individual characteristics to accurately determine the areas of territory under Ukraine's control, those under occupation prior to February 24, 2022, and areas newly occupied during the period of this study. In those instances where pixel colour data on borders made filter attribution impossible, these were classified as "others" and allocated to the newly occupied and Ukraine-controlled territory on a proportional basis. The same methodology was applied to a small number of pixels for which the filter had automatically attributed the colour to both territories. The variable is available in the news dataset.
applying a log plus one transformation to all news variables and re-estimate Eq. (3). By the same token, this robustness check would account for the fact that multiple news in one category on the same day may be linked or relate to the same event, such that weighting these down might be appropriate. Table A4.6-A4.10 show the results. Interestingly, all structural breaks (including the one for S&P 500) are now found in early-to mid-April, that is, right after the liberation of Northern Ukraine. In addition, the general pattern for the "Western support" variables is replicated when accounting for potential non-linearities. Markets appreciated a more dovish stance during the first phase of the war and a more hawkish stance later on.
To summarize, our results are robust to an alternative timing of the war-related news, using a broader European stock index, and potential non-linearities in the markets' response to news.
We also show that it is news about sanctions that matters for markets and not the number of actually sanctioned Russian entities. The same holds for news about combat escalation as compared to actually observed territorial gains or losses.

Conclusions
Since Russia began the war against Ukraine on February 24, 2022, the West has been intensively discussing its support strategy. Hawkish positions of strengthening Ukraine via armaments, financial resources, and sanctions against Russia compete with dovish views of avoiding further escalation of the military and geopolitical conflict. News from the war became a dominating factor for international politics and the world economy.
In the underlying study, we created a comprehensive data set of news related to the war in Ukraine. We focus on decisions of the West, that is, military aid, financial aid, and sanctions.
Further news categories comprise combat escalation, the export situation, and the local situation.
Building on this data, we measure reactions of international financial markets to this news. We cover five key financial markets, namely US and European stocks, oil and gas for the energy dimension, and wheat representing agriculture.
The results show that stronger support for Ukraine had a positive impact after the first weeks of the war when the Ukrainian position in the war improved, but a negative or at least less positive influence before. Thus, financial markets seem to have perceived support as a risk of further escalation threatening global economic activity in the first phase. However, a hawkish line was a positive signal for financial markets after the change in perceptions. This means that clear support for Ukraine has been viewed as stabilizing business prospects and relaxing the situation in strained energy and commodity markets. Importantly, the results also confirm that the war and escalation in general had harmful effects on international financial markets.
Our results are robust to an alternative timing of the war-related news and potential nonlinearities in the markets' response to these. We also show that it is news about sanctions that matters for markets and not the number of actually sanctioned Russian entities. The same holds for news about combat escalation as compared to actually observed territorial gains or losses.
Finally, our paper provides novel insights with respect to time-variation in asset price responses to news. While the existing literature finds such variation due to business cycles, financial market turbulences, and other secular trends (cf. footnote 1), we additionally provide evidence for the market perception of conflicts in this context.
Annex 1: The procedure for news item categorization Notes: After classifying each news story separately, we do a double-checking process and review each set of news items for a separate sectoral code and direction to ensure these belong to the same group. Notes: In each case, we access not only a headline, but a text within the news item, which was not represented in the tables above for the sake of brevity.

Annex 2: The course of the war in maps
To follow the course of the war in maps, we use the @War Mapper as a source of daily updates.
The information on which the mapping is based is collected from open sources and provides an approximation of the course of the war.
Russia invaded Ukraine on February 24, 2022. The invasion begun with missile strikes on the entire country. The ground offensive was rapid, and within less than a month Russian troops occupied parts of the Sumy, Chernihiv, and Kyiv regions in the North, Kharkiv, Luhansk, and Donetsk regions in the East, and Zaporizhzhia and Kherson regions in the South. Thereby, Russian forces have occupied the large cities of Kherson and Melitopol, surrounded and captured Mariupol after its 86 days of resistance, and got within several kilometres distance to Kyiv and Kharkiv -two of Ukraine's biggest cities as shown in Figure A2.1.

Figure A2.1: Situation on the ground as of March 19, 2022
Source: @war_mapper After the initial shock, Ukrainian defence improved and then the situation slowly started to reverse. On April 1, the War Mapper reported that Ukraine has retaken a significant number of settlements to the east and west of Kyiv, which is shown by the green areas ( Figure A2.2). Source: @war_mapper By April 4, Russian troops have been pushed out of the Kyiv region and withdrawn quickly from the Sumy and Chernihiv regions, completely leaving the North as shown in Figure A2.3.
On April 13, the Russian flagman warship Moskva sank, which was even more symbolic given that this ship threatened the Ukrainian defenders during the Russian attack on Snake Island in Ukraine's territorial waters. However, the Russian offensive in the East and South continued, and Ukraine was defending its positions against the prevailing troops of the aggressor.   Closer to the end of the year, being unable to advance on the ground, Russia has changed its strategy to air strikes against civilian targets, such as the Ukrainian electricity infrastructure.
Whereas the maps show no significant progress of each side, Russian drones and missiles have damaged or destroyed approximately 30-50 percent of Ukrainian electricity distribution capacities since mid-October, leaving the population and businesses without electricity and also without heating and water supply during the winter months.
This is the most recent update on the situation as of end-December 2022 ( Figure A2.5). The source of the maps is https://mobile.twitter.com/war_mapper.

Annex 3: Background information on Western aid and sanctions and the economic situation in Ukraine
Sanctions. Days after the start of Russia's invasion, many Russian entities and individuals got sanctioned by the EU, US, and other authorities. Russian banks got banned from SWIFT, companies, in particular those connected to the airspace or military industry, faced trade restrictions, maritime cargo traffic suffered from sanctions on cargo ships, and assets of Russian oligarchs and politically-exposed persons were seized or frozen. Further waves of sanctions followed shortly thereafter. Figure A3.1 shows the number of Russian entities being sanctioned by Western authorities. Figure A3.2 shows -for comparison -the frequency of sanctionsrelated news in our dataset.  Ukraine has survived economically and financially thanks to the lifeline provided by its Western partners. Initially, the financial assistance was not regular and was not enough to cover the budget gap. Therefore, a large share of expenditures was financed by the central bank,    On July 22 (i.e. after the sample period of this paper), the so-called "Grain Corridor" agreement was signed in Turkey, allowing maritime transportation of Ukrainian grain. The (huge) drops in exports due to the war are also visualized in Figure A3.4 above.   Notes: Table shows     (3) with heteroskedasticity and autocorrelation-robust standard errors in parentheses, controlling for the cumulative number of sanctioned entities (cf. Figure A3.1). Coefficient of Log(Cumulative Sanctioned Entities) indicates the response of market returns (in pp) to a 10 percent increase in the variable. ***/**/* indicate significance at the 1%/5%/10% level, respectively. Full tables are available on request. (3) with heteroskedasticity and autocorrelationrobust standard errors in parentheses. ***/**/* indicate significance at the 1%/5%/10% level, respectively. (3) with heteroskedasticity and autocorrelation-robust standard errors in parentheses. Coefficients of war-related news indicate the response of market returns (in pp) to a 10 percent increase in the variables. ***/**/* indicate significance at the 1%/5%/10% level, respectively. (3) with heteroskedasticity and autocorrelation-robust standard errors in parentheses. Coefficients of war-related news indicate the response of market returns (in pp) to a 10 percent increase in the variables. ***/**/* indicate significance at the 1%/5%/10% level, respectively. (3) with heteroskedasticity and autocorrelation-robust standard errors in parentheses. Coefficients of war-related news indicate the response of market returns (in pp) to a 10 percent increase in the variables. ***/**/* indicate significance at the 1%/5%/10% level, respectively. (3) with heteroskedasticity and autocorrelation-robust standard errors in parentheses. Coefficients of war-related news indicate the response of market returns (in pp) to a 10 percent increase in the variables. ***/**/* indicate significance at the 1%/5%/10% level, respectively. (3) with heteroskedasticity and autocorrelation-robust standard errors in parentheses. Coefficients of war-related news indicate the response of market returns (in pp) to a 10 percent increase in the variables. ***/**/* indicate significance at the 1%/5%/10% level, respectively.