POLICY NEWS AND STOCK MARKET VOLATILITY

Transcription

1 POLICY NEWS AND STOCK MARKET VOLATILITY Scott R. Baker Kellogg School of Management Nick Bloom Stanford University Steven J. Davis Booth School of Business, University of Chicago Kyle Kost University of Chicago March, 219 Working Paper No. 19-9

2 Policy News and Stock Market Volatility Scott R. Baker, a Nicholas Bloom, b Steven J. Davis c and Kyle Kost d 26 October 218 Abstract: We create a newspaper based tracker of Equity Market Volatility (EMV), counting the frequency of newspaper articles containing key terms about the (E)conomy, the Stock (M)arket and (V)olatility. This EMV index has a monthly correlation with realized S&P 5 volatility and the VIX of.8 and.77 respectively. Using this index we find three key results. First, 74 percent of EMV articles mention the Macroeconomic Outlook, 44 percent mention Commodity Market while 3 percent mention referencing Interest Rates. The frequency of Macroeconomic articles surges during recessions while Commodities articles are strongly linked to oil prices, with interest rates mentions relatively constant over time. Second, in terms of policies the major drivers of EMV are Fiscal Policy, Monetary Policy and National Security. More recently, since the election of Donald Trump in the number of trade policy articles has also increased sharply. Finally, the share of EMV articles containing any policy term has been rising since the 198s, suggesting policy is playing an increasing role in driving stock-market fluctuations. JEL No. D8, E22, E66, G18, L5 Keywords: stock market, volatility and uncertainty Acknowledgements: We thank the National Science Foundation, the Sloan Foundation, and the Becker Friedman Institute, Initiative on Global Markets and the Stigler Center at the University of Chicago for financial support. a Kellogg School of Management; s-baker@kellogg.northwestern.edu b Stanford; nbloom@stanford.edu c University of Chicago Booth School of Business and the Hoover Institution; steven.davis@chicagobooth.edu d University of Chicago; kkost@uchicago.edu 1

3 1. Introduction While it is easy to measure stock-market volatility using stock returns data back to before 19, determining the factors driving this is harder. Stock market volatility itself can be correlated with other factors like interest rates or commodity prices, but it is hard to determine causal relationships. Moreover many potential drivers like government policies or risks of foreign war have no obvious high-frequency measure. So as one step towards understanding the drivers of stock market volatility we use newspapers. First, we build an Equity Market Volatility (EMV) index to proxy monthly US stock market volatility. Armed with this index we then investigate what factors words, concepts and policies appear to be linked with fluctuations in US stock market volatility by examining the underlying articles that are driving our EMV index. In particular, we use the eleven major daily US newspapers 1 : the Boston Globe, Chicago Tribune, Dallas Morning News, Houston Chronicle, Los Angeles Times, Miami Herald, New York Times, San Francisco Chronicle, USA Today, Wall Street Journal, Washington Post. For these papers we constructed an index of article frequency that mentions key Equity, Market and Volatility words that mimics the VIX index of stock-market implied volatility over our training period of -. 2 Figure 1 plots this EMV index, which displays clear spikes around major periods of stock-market turbulence, so provides a reasonable measure of US stock market turbulence. In Figure 2, we plot the EMV index alongside the VIX and these clearly move closely together, albeit with some deviations around large jumps in the late 198s (after Black Monday) and the Global Financial Crisis in -. But overall the correlation with the VIX is high with a monthly of.78. Armed with this index of US stock market volatility we then proceed to evaluate what drives these periods of stock-market turbulence. The advantage of a text based index is that we can drill 1 These 11 papers are the we originally used in Baker, Bloom and Davis () plus the Houston Chronicle which is a large circulation high quality newspaper with full-text available online. 2 The standard CBOE VIX index starts in and is a measure of equity market volatility, so when we started this project we had VIX data from -. Subsequently we obtained VIX data developed in Berger et al. (218) back to

4 down into the individual articles generating the fluctuations in equity market volatility to determine what factors newspaper attribute this too. We find three key sets of results. First, 74 percent of EMV articles make remarks about the Macroeconomic Outlook, 44 percent mention Commodity Market while 3 percent referencing Interest Rates. This highlights how macroeconomic factors are the major driver of US stock-market fluctuations, particularly during recessions when the frequency of Macroeconomic surges. Commodities articles primarily focus oil so are strongly linked to oil price volatility, which in turn is strongly linked to oil price levels. Finally, interest rates mentions tend to be relatively more constant over time, perhaps highlighting the more stable influence of the Federal Reserve System. Second, in terms of policies the major drivers are Fiscal Policy, Monetary Policy and National Security. National Security spikes during periods of war for example the Gulf Wars and after the 9/11 terrorist attacks. More recently since the election of Donald Trump in the number of trade policy articles has also increased sharply. Third, the share of EMV articles containing any policy term has been risen by around a quarter since the 198s. This suggests that policy may be playing an increasing role in driving stockmarket fluctuations, highlighting the trend found in terms of a rising ratio of the Economic Policy Uncertainty index to the VIX reported in Baker, Bloom and Davis (). This paper is linked to four prior literatures. One focused on evaluating the drivers of stockmarket fluctuations. Cutler, Poterba and Summers (). Pastor and Veronesi (, ) find government policy uncertainty to be a strong driver of stock prices and risk premia. Baker, Bloom, Davis, and Sammon (218) show that government policy is also a big driver of large stock market movements through an analysis of text in the days after these large market swings. Our work directly measures the importance of government policy in stock market volatility and sheds light on the increasing role policy may be having over time. Second, another literature details how stock market volatility as measured by the VIX is useful in predicting other financial variables of importance. Dreschler and Yaron () show that the variance premium as measured by the difference between the squared VIX and expected realized variance displays significant time variation and return predictability. Nagel () shows that the return from liquidity provision is highly predictable with the VIX. The EMV index and its 3

5 related indices developed here can be used to better understand these relationships as we can dig deeper into what drives the VIX and stock market volatility in general. A third literature uses newspaper and other text sources to try and model financial markets. The most closely related paper in this vein is Manela and Moreira () who develop their NVIX measure of stock market volatility using the text from front pages of the Wall Street Journal. Their methodology differs significantly though as they only use one section of one newspaper and utilize machine learning techniques to derive their index. Hassan et al. (218) also use machine learning techniques on company conference call transcripts to develop companyspecific measures of political risk that they show drive lobbying expenditures. Our approach instead more closely follows that of Baker, Bloom, and Davis () who develop a measure of economic policy uncertainty using a similar set of newspapers. Finally, as text sources are further explored as potential data sources, much work is being developed in refining the tools to analyze text. Recent papers doing this are Gentzkow, Kelly, and Taddy (217) and Kelly, Manela, and Moreira (218). The first explores summarize recent papers using text-related methods and the tools that are used in each while the second improves upon existing high dimensional text selection models and shows the gains from the new methodology via backcasting financial variables to historical periods. Our methodology highlights a simple alternative to complicated machine learning algorithms to explore text as data. Section 2 discusses our methodology, section 3 validates our equity market volatility index, section 4 reports our key results while section 5 concludes. 2. Methodology Our Equity Market Volatility Index currently covers the period -217 (but will be updated monthly on We construct this index as follows. First, we build a term set for topics which for Equities is [economic OR economy OR financial], for Stock Markets is [ stock market OR equity OR equities OR Standard and Poors OR Standard & Poors OR Standard and Poor OR Standard and Poor's OR Standard & Poor s ], and for Volatility is [uncertain OR uncertainty OR volatility OR volatile OR risk OR risky]. This term 4

6 set was chosen after a period of trial and error with a set of plausible terms related to equity market volatility. While this did involve attempting to improve the fit to the VIX this was not an exhaustive machine learning data-fitting exercise, and instead involved testing the fit of a relatively narrow set of terms. In particular, we wanted to avoid picking terms like housing crisis, Lehman Brothers, Bernanke or Gulf war that could have produced a good insample fit for stock-market volatility but would likely have limited future predictive power. Armed with this term set we generate monthly counts of articles in each of our 11 national newspapers. These are the Boston Globe, Chicago Tribune, Dallas Morning News, Houston Chronicle, Los Angeles Times, Miami Herald, New York Times, San Francisco Chronicle, USA Today, Wall Street Journal, Washington Post. We chose these are defined on being regional or national daily US newspapers which are available in full-text online (so we can scrape them). The frequency of these terms are then scaled by the total number of articles in the newspaper for that month. For each newspaper, we standardize the resulting time series to have unit standard deviation over the - time period. Finally, we average these newspaper level indices across newspapers at the monthly level and rescale the final index to have the same mean as our extended VIX measure over -. Five of the newspapers we use do not have full coverage for the -217 time period: the Dallas Morning News, Houston Chronicle, New York Times, USA Today, and Washington Post. Access World News has stopped covering the Dallas Morning News, so it is no longer being updated for July and beyond. Thus, coverage for us stops in the middle of. The ProQuest newspaper archive only covers the New York Times through currently. 3 The USA Today from Access World News is missing coverage from,, and the first half of, the Houston Chronicle is missing coverage for through October, and the Washington Post from ProQuest is missing coverage in and. We choose to impute missing scaled counts for these five newspapers. We impute missing values using the fitted values from the following regression: 3 The New York Times gets updated with some delay on ProQuest. All of the imputation, standardization, and normalization is done with the New York Times data up through as that is what is available at the time of writing. 5

7 DMNBt = β+ N* βi Newspapersit + εit where N* is the set of newspapers for which we have complete coverage (Boston Globe, Chicago Tribune, Los Angeles Times, Miami Herald, San Francisco Chronicle, and Wall Street Journal). DMNBt is the scaled monthly counts for the Dallas Morning News, and each of the regressions is over the - time period. The same imputation is done for the other newspapers missing values. We fix the imputation coefficients from these regressions for future imputations. 4 There are many text sources to potentially choose from when deciding to construct an index such as the one presented here. Newspapers are convenient as there are many online archives that can be easily searched and explored that have articles over a long time period, and they can be combined to represent a good geographic mix of content and discussion. There are also some potential drawbacks. One concern is that newspapers are not the primary source for news anymore with so many other online sources. Figure A1 plots the total articles from the newspapers we use for our index. We see that the total article counts remained relatively stable only with some decreases in the last decade or so. Another potential concern is exactly how newspapers are chosen for the index and how they are combined in the final index construction. We explored the composition of articles in the newspapers we use for our index to better understand this. Table A1 reports the average number of articles per day for the different topic term sets E (economy), M (stock market), and V (volatility) along with the percent of the total number of articles these groups represent. We see that for the most part each of the newspapers included in the index use the same fraction of their total coverage on the topics we look at. The one main standout is the Wall Street Journal which spends a lot of coverage on stock markets and market volatility in particular. Table A2 reports coefficients from our baseline regression of the VIX on the EMV Index except now each column represents a version of the EMV Index with a particular newspaper s weight 4 This means that as more New York Times data is available with a delay, we will not re-estimate the imputation regressions, but instead we just use the fitted values that we would get from the original coefficients. We do this to avoid updating historical data we post online. 6

8 doubled. We see that the coefficients are very similar with small changes in the R 2 values across specifications. This is even true in column (9) where we look at increasing the weight of the Wall Street Journal. 3. Validation Before proceeding to analyze the drivers of our EMV index, we need to first validate it captures the main shifts in US stock-market volatility. To some extent Figure 1 undertook an informal version of this in that the major jumps in the EMV index line-up with notable shocks to US financial market. But in this section we want to evaluate our EMV index more formally. First, we start with Figure 2 which plots the EMV against the VIX the index of 1 month ahead implied volatility on the S&P5 index. We see a reasonably good fit they have a correlation of.78 with similar peaks around major financial shocks. There are some divergences, however, most notably when VIX is above the EMV after Black Monday in, during Gulf War I in and around and 2. In reverse, during some of the periods of low stockmarket volatility in the mid-s and mid-s the VIX was persistently below the EMV. Figure A2 reports the residuals from a regression of the VIX on our EMV Index with the ten largest values labeled. As noted before, the divergences are primarily located around Black Monday in as well as in the Great Recession. Table 1 provides regression analysis of the VIX and realized volatility on our EMV index. Column (1) shows our baseline specification of the contemporaneous association of the EMV index with the VIX showing strong statistical significance and a high R 2 value of.61. Hence, in the raw data around 6% of the variation in the VIX can be matched by our newspaper EMV index. In column (2) we add one and two lags of the EMV index and the r-squared rises to.7 highlighting how the lagged EMV index is also informative. Column (3) adds a square of the EMV index with no improvement in fit, while column (4) runs a log-log specification again with no improvement in fit, collectively suggesting the relationship is approximately linear 5. 5 We experimented with a number of other nonlinearities and did not find anything that notably improved the fit. 7

9 In column (5) we include lagged VIX and find this substantially improves the fit, pushing the r- squared up to.84 highlighting that the gaps between the VIX and EMV are auto-correlated as shown in Figure 2. Column (6) added lagged EMV and shows this is no longer significant after controlling for lagged VIX. Column (7) includes both lagged EMV and VIX so putting them on the same timing (averages over the prior month) to see which does the better job of forecasting current VIX. It is clear that lagged VIX is a far superior predictor of current VIX, with the EMV showing a positive but insignificant coefficient, suggesting it provides at best only weak marginal additional predictive power for VIX. Finally, columns (8) to () repeat some of the key regressions using realized volatility rather than VIX, and display similar findings. Interestingly, the fit to realized volatility is slightly better (e.g. r-squared of.65 in column (8) versus.61 in column (1)). This is not surprising newspaper coverage of stock markets tends to focus on realized volatility, with articles on major stockmarket crashes or rallies, with the correlation with VIX presumably because this is also highly correlated with realized volatility. Table 2 compares our EMV index to an alternative news-based VIX index, the NVIX. The NVIX uses a support vector machine approach on the front page titles and abstracts of WSJ articles machine learning approach. In summary, this looks for one and two word combinations that over their training sample period of to best fit the VXO index 6. Examples include words like stock, market, war, US, Tax, Special, Washington and Banks. So, compared to our approach this has the advantage of automatically picking terms based on a systematic computer search, which should substantially improve in-sample fit. It has the downside of being based on a narrower body of text (just the front-page of the WSJ), and potentially overfitting the data if words like Tax or Banks were important for stock volatility from to (a period dominated by the global financial crisis) but not in earlier or later periods. So to evaluate this table is broken down by time period with Panel A covering the whole overlapping sample of -March for which NVIX and EMV are both available, Panel B covering the training sample period for the NVIX -, and Panel C covering the NVIX pre-period - (this was the test period they used in their paper), and the Panel D 6 The VXO index was the predecessor to the VIX index generated by the CBOE, and has a correlated with the VIX of.98 between to. 8

10 covering the post-period of 2-March. We see in Panel A that our EMV index matches the VIX more closely across the summary statistics shown: standard deviation, skewness, and kurtosis, with an entire sample period of.78 compared to.7 for the NVIX. In panel B we see as expected the NVIX performs better in terms of fit during its sample period of -. In panels C and D we see in the prior and post NVIX training periods the EMV has a higher match to the VIX. The main take-away being that the machine learning techniques for text analysis used in developing the NVIX can be powerful for in-sample fitting but potentially perform worse out-ofsample because the lack of economic content used to pick the terms. Of course, the acid key test for EMV against the NVIX will be on future data from 219 onwards when both indices will be out of sample. Table A3 in the appendix reports regressions of the VIX on the EMV and NVIX indices. We see that the EMV Index generally has a higher R 2 value and is still statistically significant in the contemporaneous regression even with the NVIX included. Figure A3 plots the VIX, NVIX, and EMV indices form through March. We can see the improved fit the EMV Index provides in the earlier and later parts of the sample. 4. What Drives Stock Market Volatility Having established our EMV index does a reasonably good job of matching the VIX implied volatility and realized volatility S&P 5 measures we proceed to evaluate what drives its variations. We start by using the category-specific term sets included in the appendix to determine the percentage of articles that have one or more terms related to a given policy category. We do this analysis on the full EMV Index from Table 3 reports the total percentages for the full time period for each policy category while Table 4 breaks this category level analysis down into different time periods. 9

11 We see from Table 3 that the main non-policy drivers of our index are Macroeconomic News and Outlook (72%), Commodities (43%), and Interest Rates (31%). Thus, macroeconomic factors appear to be the major drivers of US stock market fluctuations, followed by commodities which mainly reflects oil prices with some less frequent mentions of gold, with interest rates mainly reflecting Federal Reserve policy. In the lower panel of Table 3 we compare the frequencies of topics within the EMV index with the Economic Policy Uncertainty index of Baker, Bloom and Davis (). Mostly the factors that drive policy uncertainty fiscal and monetary policy and regulation play a role in driving stock-market volatility. This highlights how policy uncertainty and stock-market volatility appear to be strongly intertwined a topic we will return to below. The only major areas of divergence are around financial regulation - which is more than twice as relevant for the EMV as for the EPU index and government spending, entitlement programs and national security, which are about twice as influential for the EPU index. In Table 4 we examine our major categories over time and see a few key results. First, not surprisingly, there is a considerable increases in the role of Macro News and Outlook during the early s and the Great Recession. On the other hand, Interest Rate discussion is more stable overtime potentially alluding to the stability of the Federal Reserve system in both good times and bad. Second, National Security spikes in periods of war such as the Gulf Wars with an almost three-fold increase relative to its overall average after the 9/11 terrorist attacks. Fiscal Policy, Monetary Policy, and Regulation (specifically, Financial Regulation) increase during the Great Recession. More recently, since the election of Donald Trump in, Trade Policy has increased by a factor of 2 relative to its overall average. 4.1 Policy-Related EMV Index We also construct an economic policy market volatility index derived from policy related discussion in our EMV index. For each newspaper we calculate the share of EMV articles that mention one of the policy categories shown in Table 3. We then average these ratios across the set of newspapers N by month to get the corresponding monthly Policy Ratio time series

12 Policy Ratiot = 1/N Policy Ratio, Figure 3 plots the fraction of EMV articles that contain policy-related terms. We see large swings in the role of policy for stock market volatility, but also a clear upwards trend highlighting the increasing role policy has had on stock market fluctuations. Starting in with a policy share of a little less than.4, we are currently hovering near.5 consistently in recent years. This rise in the policy share has been partially driven by big policy-related events in the s including the Gulf Wars, debt-ceiling debate, and fiscal cliff, and then more recently with Brexit and the election of Donald Trump and his stance on trade policy among other issues. Figure A4 plots the fraction of EMV articles that contain policy-related terms from the policy term set used in Baker, Bloom, and Davis (). We see that the overall pattern remains the same with a slight level shift up in the fraction when using this other term set. In particular, we still see an increasing trend over the entire period. Figure 4 plots the Policy-Related EMV Index which is defined as the total EMV index scaled by the policy share and again renormalized to match the mean of the Baker, Bloom, and Davis () EPU Index from -. Clearly these indices have their differences suggesting the role for the Policy-Related EMV Index as another unique measure of policy uncertainty. In particular, there are a few time periods where the two indices diverge. In the early s and more recently right after the financial crisis and with the election of Donald Trump, we see higher levels of Economic Policy Uncertainty than the Policy-Related Equity Market Volatility. These episodes contain a great deal of policy uncertainty that did not translate into as much stock market volatility possibly because of the influence of other market calming forces. On the other hand, in the late s and early s, we see more Policy-Related Equity Market Volatility than Economic Policy Uncertainty with the Russian Financial Crisis and Dot-Com Crash. These distinctions emphasize a need to better understand how policy uncertainty can directly translate into stock market fluctuations and which policy categories are more likely to influence markets. 5. Conclusion We create a newspaper based tracker of Equity Market Volatility (EMV), counting the frequency of newspaper articles containing key terms about the (E)conomy, the Stock (M)arket and 11

13 (V)olatility. This EMV index has a monthly correlation with realized S&P 5 volatility and the VIX of.8 and.78 respectively. Using this index we find three key results. First, 74 percent of EMV articles mention the Macroeconomic Outlook, 44 percent mention Commodity Market while 3 percent mention referencing Interest Rates. The frequency of Macroeconomic articles surges during recessions while Commodities articles are strongly linked to oil prices, with interest rates mentions relatively constant over time. Second, in terms of policies the major drivers of EMV are Fiscal Policy, Monetary Policy and National Security. More recently, since the election of Donald Trump in the number of trade policy articles has also increased sharply. Finally, the share of EMV articles containing any policy term has been rising since the 198s, suggesting policy is playing an increasing role in driving stock-market fluctuations. 12

14 References Alizadeh, Sassan, Michael W. Brandt, and Francis X. Diebold,. Range-Based Estimation of Stochastic Volatility Models, Journal of Finance, 57, Andersen, Torben G., Tim Bollerslev, Francis X. Diebold, and Clara Vega,. Micro Effects of Macro Announcements: Real-Time Price Discovery in Foreign Exchange, American Economic Review, 93, Andersen, Torben G., Tim Bollerslev, Francis X. Diebold, and Clara Vega,. Real-Time price discovery in global stock, bond, and foreign exchange markets, Journal of International Economics, 73, Bachmann, Rüdiger, Steffen Elstner, and Eric R. Sims,. Uncertainty and Economic Activity: Evidence from Business Survey Data, American Economic Journal: Macroeconomics, 5, 2, Baker, Scott, Nicholas Bloom and Steven J. Davis,. Measuring Economic Policy Uncertainty, Quarterly Journal of Economics, doi:.93/qje/qjw24. Baker, Scott, Nicholas Bloom, Steve J. Davis and Marco Sammon, 218. What triggers stock market jumps? Work in progress presented at the January 218 ASSA meetings. Berger, David, Ian Dew-Becker, and Stefano Giglio, 218. Uncertainty shocks as secondmoment news shocks, Working paper. Christensen, B.J. and N.R. Prabhala,. The relation between implied and realized volatility, Journal of Financial Economics, 5, 2, Cutler, David M., James M. Poterba, and Lawrence H. Summers,. What moves stock prices? Journal of Portfolio Management, 15, 3, 4-12 Drechsler, Itamar and Amir Yaron,. What s Vol Got to Do with it?, The Review of Financial Studies, 24, 1, Gentzkow, Matthew, Bryan T. Kelly, and Matt Taddy, 218. Text as data, Journal of Economic Literature (Forthcoming). Gorodnichenko, Yuriy and Michael Weber,, Are Sticky Prices Costly? Evidence from the Stock Market, American Economic Review, 6, 1, Jurado, Kyle, Sydney Ludvigson, and Serena Ng,. Measuring Uncertainty, American Economic Review, 5 (3), Kelly, Bryan, Asaf Manela, and Alan Moreira, 218. Text Selection, Work in progress. 13

15 Londono, Juan M. and Beth Anne Wilson, 218. Understanding Global Volatility, IFDP Notes, January 218. Manela, Asaf and Alan Moreira, 217. News implied volatility and disaster concerns, Journal of Financial Economics, 123: Nagel, Stefan,. Evaporating Liquidity, Review of Financial Studies, 25, 7, Pastor, Lubos and Pietro Veronesi,. Uncertainty about Government Policy and Stock Prices, Journal of Finance, 64, 4, Pastor, Lubos and Pietro Veronesi,. Political Uncertainty and Risk Premia, Journal of Financial Economics, 1, 3, Appendix The following is a list of the policy/non-policy categories that we use to decompose the EMV Index in Tables 3 and 4 along with the term sets we use to identify an article from that category. The decomposition term sets for each category are then constructed by appending the terms listed for the corresponding category to the original EMV Index term set. As an example, the following would be our search term set for the Food and Drug Policy decomposition series: {economic OR economy OR financial} AND {uncertain OR uncertainty} AND { stock market OR equity OR equities OR S&P OR S & P } AND { prescription drug act OR drug policy OR food and drug administration OR fda}. The Regulation and Fiscal Policy categories are further broken down into numerous subcategories. For the Fiscal Policy All Subcategories and Regulation All Subcategories categories, we use all of the fiscal policy and regulation subcategory search terms respectively in our query string. Unfortunately, the query string for the Regulation All Subcategories category is too long for the search interfaces, and therefore we cannot scrape that series. Instead, we construct a raw Regulation category that uses the search terms {regulate, regulation, regulatory} as listed next to All Regulation below. Other Categories and Topics Macroeconomic News and Outlook: {gold, silver}, {gdp, economic growth}, {depression, recession, economic crisis}, {macroeconomic indicators, macroeconomic news}, {rail loadings, railroad loadings}, {cpi, inflation, consumer prices, ppi, producer prices}, {housing prices, home prices, homebuilding, homebuilders, housing starts, home sales, building permits, residential sales, mortgages, residential construction, commercial construction, commercial real estate, real estate}, {labor force, workforce, unemployment, employment, unemployment insurance, ui claims, jobs report, jobless claims, payroll, underemployment, 14

16 quits, hires, weekly hours}, {wages, labor income, labor earnings}, {corporate bonds, bank loans, interest rates}, {trade news, trade surplus, trade deficit, national exports, national imports}, {business investment, business inventories}, {consumer spending, retail sales, consumer purchases}, {consumer confidence, consumer sentiment, macro outlook, business sentiment, business confidence}, {industrial production, ism report, manufacturing index},, {household credit, household savings, household debt, household borrowing, consumer credit} Commodities: {wheat, corn, soy, sugar, cotton, beef, pork}, {oil, petroleum, coal, natural gas}, {biofuel, ethanol}, {steel, copper, zinc, tin, platinum, rare earth metals, gold, metal, silver, aluminum, lead}, {cme, commodity exchange, cbot, nymex, lme, London metal exchange, mercantile exchange, intercontinental exchange, board of trade}, {keystone pipeline, Alaska pipeline, gas pipeline} Interest Rates: {interest rates, yield curve, fed funds rate, overnight rate, repo rate, T-bill rate, bond rate, bond yield} Financial Crises: {financial crisis, financial crises}, {Northern Rock failure, Lehman failure, Lehman Brothers failure, AIG Takeover}, {euro crisis, Eurozone crisis, Greek crisis} Exchange Rate: {exchange rate}, {currency crisis}, {currency devaluation, currency depreciation}, {currency revaluation, currency appreciation}, {crawling peg, managed float}, {currency manipulation, currency intervention} Healthcare Matters: {healthcare}, {health insurance}, {Medicaid}, {Medicare}, {Affordable care act, Obamacare}, {medical liability, medical malpractice}, {prescription drug}, {drug policy}, {food and drug administration, fda}, {VA hospital, VA healthcare, Veterans Affairs hospital, Veterans Affairs healthcare, Veterans Health Administration}, {National Institutes of Health} Litigation Matters: {lawsuit, litigation, class action, tort}, {punitive damages}, {patent infringement, trademark infringement, copyright infringement}, {medical malpractice}, {Supreme Court} Competition Matters: {antitrust, competition policy, competition law}, {federal trade commission, ftc}, {unfair business practice}, {monopoly, monopolization}, {cartel}, {price fixing, price conspiracy}, {Sherman Act}, {Robinson Patman Act}, {Clayton Act}, {Hart- Scott-Rodino}, {European Commission} Labor Disputes: {labor dispute, labor unrest, strike}, {labor litigation, employee discrimination, wage and hour litigation, labor class action} Intellectual Property Matters: {patent}, {trademark}, {copyright}, {Patent and Trademark Office}, {International Trade Commission}, {federal trade commission, ftc}, {intellectual property}, {Hatch-Waxman}, {new drug application} Policy Categories and Policy Topics All Taxes: {taxes, tax, taxation, taxed}, {income tax, tax on individuals, personal tax}, {capital gains tax, tax on capital gains}, {dividend tax}, {mortgage interest deduction, deduction for mortgage interest}, {IRA account, Roth IRA, traditional IRA, 41-k}, {state and local tax deduction, deductibility of state and local tax}, {payroll tax, social security tax, social security contributions, Medicare taxes, FICA, unemployment tax, FUTA}, {sales tax, excise tax, value added tax, vat, goods and services tax, gross receipts tax}, {carbon tax, 15

17 energy tax}, {corporate tax, business tax, profit tax}, {investment tax credit, accelerated depreciation}, {R&D tax credit, research and development tax credit}, {tax credit for lowincome housing, low-income housing credit}, {black liquor tax credit, black liquor credit}, {ethanol credit, ethanol credit, ethanol tax rebate}, {biofuel tax credit, biofuel producer tax credit, fuel excise tax rebate, fuel tax credit, alcohol fuel credit}, {property tax}, {fiscal cliff}, {Internal Revenue Service} Government Spending, Deficits and Debt: {government spending, government outlays, government appropriations, government purchases}, {defense spending, military spending, defense purchases, military purchases, defense appropriations}, {entitlement spending}, {government subsidy}, {fiscal stimulus}, {government deficit}, {federal budget, government budget}, {Gramm Rudman, balanced budget, balance the budget, budget battle, debt ceiling}, {fiscal cliff, government sequester, budget sequestration, government shutdown}, {sovereign debt}, Entitlement and Welfare Programs: {entitlement program, entitlement spending, government entitlements}, {social security, Supplemental Security Income, ssi, disability insurance}, {Medicaid}, {Medicare}, {supplemental nutrition assistance program, food stamps, wic program}, {unemployment insurance, unemployment benefits, TAA program}, {welfare reform, aid to families with dependent children, afdc, temporary assistance for needy families, tanf, public assistance}, {earned income tax credit, eitc}, {head start program, early childhood development program}, {affordable housing, section 8, housing assistance, government subsidized housing} Fiscal Policy: Use all of the terms from the following subcategories o Taxes: see above o Government Spending, Deficits and Debt: see above o Entitlement and Welfare Programs: see above Government-Sponsored Enterprises and Related Agencies: {Federal Home Loan Mortgage Association, Freddie Mac}, {Fannie Mae, Federal National Mortgage Association}, {Federal Housing Finance Agency}, {Federal Housing Agency}, {Sallie Mae, Student Loan Marketing Association}, {Government National Mortgage Association, Ginnie Mae}, {Federal Home Loan Bank}, {Federal Farm Credit Bank, Federal Agricultural Mortgage Corporation, Farmer Mac}, {Resolution Funding Corporation, REFCORP} Monetary Policy: {monetary policy}, {money supply, open market operations}, {fed funds rate}, {discount window}, {quantitative easing}, {forward guidance}, {interest on reserves}, {taper tantrum}, {Fed chair, Greenspan, Bernanke, Volker, Yellen, Draghi, Kuroda, Jerome Powell}, {lender of last resort}, {central bank}, {federal reserve, the fed}, {European Central Bank, ecb}, {Bank of England}, {bank of japan}, {people s bank of china, pboc, pbc, central bank of china}, {Bank of Italy}, {Bundesbank} Regulation: {regulation, regulatory, regulate} o Financial Regulation: {bank supervision}, {thrift supervision}, {financial reform}, {truth in lending}, {firrea}, {Glass-Steagall}, {Sarbanes-Oxley}, {Dodd-frank}, {tarp, Troubled Asset Relief Program}, {Volcker rule}, {Basel}, {capital requirement}, {stress test}, {deposit insurance, fdic}, {federal savings and loan insurance corporation, fslic}, {office of thrift supervision, ots}, {comptroller of the currency, occ}, {commodity futures trading commission, cftc}, {Financial Stability Oversight Council}, {house financial services committee}, {securities and exchange 16

18 commission, sec}, {Bureau of Consumer Financial Protection, Consumer Financial Protection Bureau, CFPB}, {SBA loan program} o Competition Policy: {antitrust policy, competition policy, competition law}, {federal trade commission, ftc}, {Sherman Act}, {Robinson Patman Act}, {Clayton Act}, {Hart-Scott-Rodino}, {European Commission} o Intellectual Property Policy: {patent policy, patent law}, {trademark policy, trademark law}, {copyright law}, {Patent and Trademark Office}, {International Trade Commission} o Labor Regulations: {Department of Labor}, {national labor relations board, nlrb}, {union rights, card check, right to work, closed shop}, {wages and hours, overtime requirements}, {minimum wage, living wage}, {workers compensation}, {Occupational Safety and Health Administration, osha, Mine Safety and Health Administration}, {employment at will, advance notice requirement, at-will employment}, {affirmative action, equal employment opportunity, eeoc}, {trade adjustment assistance}, {Davis-Bacon}, {ERISA}, {Pension Benefit Guaranty Corporation, PBGC} o Immigration: {immigration policy, immigration reform, migration reform}, {Immigration and Customs Enforcement, immigration and naturalization service}, {immigrant workers, immigrant labor}, {farm worker jobs program, farm worker program farm worker program, farmworker program, guest worker program, guestworker program, H-2A program, H-2B program}, {H-1B program, H-1B visa}, {refugee crisis}, {Schengen} o Energy And Environmental Regulation: {energy policy}, {energy tax, carbon tax}, {cap and trade}, {cap and tax}, {drilling restrictions}, {offshore drilling}, {pollution controls, environmental restrictions, clean air act, clean water act}, {environmental protection agency, epa}, {wetlands protection}, {Federal Energy Regulatory Commission, FERC}, {ethanol subsidy, ethanol tax credit, ethanol credit, ethanol tax rebate, ethanol mandate, biofuel tax credit, biofuel producer tax credit}, {corporate average fuel economy, CAFE standard}, {endangered species}, {Keystone pipeline}, {Alaska oil pipeline, Trans-Alaska pipeline}, {greenhouse gas regulation, climate change regulation}, {Nuclear Regulatory Commission}, {Pipeline and Hazardous Materials Safety Administration} o Lawsuit And Tort Reform, Supreme Court Decisions: {tort reform}, {class action reform}, {punitive damages reform}, {medical malpractice reform}, {lawsuit reform}, {Supreme Court} o Housing and Land Management: {Federal Housing Administration}, {Federal Housing Finance Agency}, {Department of Housing and Urban Development, HUD}, {Section 8 Housing}, {Office of Fair Housing and Equal Opportunity, FHEO}, {Bureau of Land Management}, {Department of Interior}, {zoning regulations, zoning laws}, {endangered species}, {US Forest Service, United States Forest Service} o Other Regulation: {Consumer Product Safety Commission}, {Department of Education}, {Small Business Administration}, {Federal Communications Commission, FCC}, {Fish and Wildlife Service} National Security: {national security}, {war, military conflict, military action}, {terrorism, terror, 9/11}, {defense spending, defense policy, military spending}, {Department of 17

19 Defense}, {Department of Homeland Security}, {Defense Advanced Research Projects Agency, DARPA}, {armed forces}, {base closure}, {military procurement}, {no-fly zone}, {Syrian war}, {Iraq war}, {Libyan war}, {Ukraine conflict, Ukraine invasion, Crimean invasion, Crimean annexation}, {South China Sea conflict}, {naval blockade, military embargo} Trade Policy: {trade policy}, {tariff, import duty}, {import barrier, import restriction}, {trade quota}, {dumping}, {export tax, export duty}, {trade treaty, trade agreement, trade act}, {wto, world trade organization, Doha round, Uruguay round, gatt}, {export restriction}, {investment restriction}, {Nafta, North American Free Trade Agreement}, {Trans-Pacific Partnership, TransPacific Partnership}, {Federal Maritime Commission}, {International Trade Commission}, {Jones Act}, {trade adjustment assistance} Healthcare Policy: {healthcare policy}, {health insurance}, {Medicaid}, {Medicare}, {Affordable care act, Obamacare}, {malpractice tort reform, malpractice reform}, {VA hospital, VA healthcare, Veterans Affairs hospital, Veterans Affairs healthcare, Veterans Health Administration}, {National Institutes of Health} Food and Drug Policy: {prescription drug act}, {drug policy}, {food and drug administration, fda} Transportation, Infrastructure and Public Utilities: {Department of Transportation}, {Federal Highway Administration}, {federal highway fund}, {National Highway Traffic Safety Administration}, {U.S. Surface Transportation Board}, {Amtrak, National Railroad Passenger Corporation}, {Bonneville Power Administration, Tennessee Valley Authority, Southeastern Power Administration, New York Public Power Authority, Santee Cooper, South Carolina Public Service Authority, Salt River Project, Los Angeles Department of Water and Power}, {Corps of Engineers}, {Federal Aviation Administration, FAA}, {Federal Maritime Commission}, {National Aeronautics and Space Administration, NASA}, {Pipeline and Hazardous Materials Safety Administration} Elections and Poltical Governance: {presidential election}, {Congressional election}, {parliamentary election}, {presidential impeachment}, {Brexit}, {Scottish referendum}, {Grexit, Greek exit}, {Eurozone exit, Eurozone breakup}, {military takeover, coup}, {civil war} Agricultural Policy: {Department of Agriculture, USDA}, {ethanol subsidy, ethanol tax credit, ethanol credit, ethanol tax rebate, ethanol mandate, biofuel tax credit, biofuel producer tax credit} 18

20 Table 1: Regressions of the 3-Day VIX and Realized Volatility on Our Equity Market Volatility Tracker Dep Var. EMVt EMVt-1 EMVt-2 EMVt 2 Log(EMVt) VIXt-1 RVolt-1 (1) VIX.76*** (.6) (2) VIX.53*** (.).2** (.).2*** (.6) (3) VIX 1.*** (.17) -.4 (.3) (4) VIX.43*** (.7).58*** (.8) (5) VIX.47*** (.8) -.12 (.11) -.2 (.8).67*** (.9) (6) Log(VIX).78*** (.4) (7) RVol.96*** (.9) (8) RVol.79*** (.13) R Observations Notes: All series are at the monthly level running from to 217. VIX refers to the monthly average of daily close of the VIX implied volatility index on the S&P5. RVol refers to standard-deviation of the daily returns on the S&P5 within the month. EMV is the Equity Markets Volatility Index, with subscripts indicating timing (t is contemporaneous with the dependent variable, t-1 is lagged one month etc). Newspapers used: Miami Herald, Dallas Morning News, Houston Chronicle, San Francisco Chronicle, USA Today, Washington Post, Chicago Tribune, Boston Globe, Los Angeles Times, New York Times, and Wall Street Journal. Newey-West standard errors in parentheses. * p <., ** p <.5, *** p <.1.24 (.15) 1

21 Table 2: EMV Index, NVIX, and VIX Summary Statistics Panel A Full Period: - March Summary Statistic VIX EMV NVIX Standard Deviation Skewness Kurtosis Pairwise Correlation with VIX Average Absolute Difference with VIX Panel B NVIX Sample Period: Standard Deviation Skewness Kurtosis Pairwise Correlation with VIX Average Absolute Difference with VIX Panel C NVIX Pre-Period: Standard Deviation Skewness Kurtosis Pairwise Correlation with VIX Average Absolute Difference with VIX Panel D NVIX Post-Period: 2 March Standard Deviation Skewness Kurtosis Pairwise Correlation With VIX Average Absolute Difference with VIX Notes: All series are at the monthly level. VIX refers to the monthly average of daily close of the VIX implied volatility index on the S&P5, EMV is the Equity Markets Volatility Index, and NVIX is the news-based volatility measure developed in Manela and Moreira (217) using front page articles from the Wall Street Journal. The EMV Index and NVIX were normalized to have the same mean as the VIX over the period - for this table. The NVIX is trained on the - subsample of data. Newspapers used: Miami Herald, Dallas Morning News, Houston Chronicle, San Francisco Chronicle, USA Today, Washington Post, Chicago Tribune, Boston Globe, Los Angeles Times, New York Times, and Wall Street Journal. 2

22 Table 3: Percent of EMV Articles with Terms in the Indicated Categories, -217 Panel A. Selected Categories Percent of EMV Articles Macro News and Outlook 72.3 Commodities 43.4 Interest Rates 3.7 Financial Crises 8.1 Exchange Rates 2. Healthcare Matters 6.4 Litigation Matters 4.7 Competition Matters 3.8 Labor Disputes 4. Intellectual Property Matters 3.3 Panel B. Policy-Related Categories Percent of EMV Articles Percent of EPU Articles Fiscal Policy: Taxes Government Spending, Deficits, and Debt Entitlement and Welfare Programs Monetary Policy Regulation (Generic, Financial, Competition, Labor, Lawsuit): Generic Regulation (regulate, regulation, regulatory) Financial Regulation Competition Policy Intellectual Property Policy.1.3 Labor Regulations Immigration Energy and Environmental Reform Lawsuit and Tort Reform, Supreme Court Decisions Housing and Land Management Other Regulation Education, Communications, Product Safety, Wildlife, Small Business National Security Policy Government-Sponsored Enterprises (e.g. Fannie Mae) Trade Policy Healthcare Policy Food and Drug Policy Transportation, Infrastructure, and Public Utilities Elections and Political Governance Agricultural Policy.2.6 Notes: The baseline EMV Index term set is (economic OR economy OR financial) AND (uncertain OR uncertainty OR volatility OR volatile OR risk OR risky) AND ( stock market OR equity OR equites OR Standard and Poors OR Standard & Poors OR Standard and Poor OR Standard and Poor's OR Standard & Poor s ). The baseline EPU Index term set is (economic OR economy) AND (uncertain OR uncertainty) AND (congress OR deficit OR Federal Reserve OR the Fed OR legislation OR legislative OR legislature OR regulation OR regulatory OR white house ). The category specific indices additionally 3

23 contain a category specific term set. See the appendix for the full details of the category-specific termsets. corresponds to a percent that is exactly equal to whereas. corresponds to one that rounded to zero. Newspapers used: Miami Herald, Dallas Morning News, Houston Chronicle, San Francisco Chronicle, USA Today, Chicago Tribune, Boston Globe, Washington Post, Wall Street Journal, New York Times, and Los Angeles Times. 4

24 Table 4: Equity Market Volatility by Policy Category and Time Period, to 217 Time Period :1- :12 :1- :12 :1 - :8 Boom :9- :12 5 :1 - :6 s :7 - :8 Credit :9- :12 2:1- :12 214:1- : :11-217:12 :1-217:12 Gulf War I to 9/11 9/11 Attacks boom Crunch Lehman/ recession Policy Battles Equity Market Volatility Macro News and Outlook Commodities Interest Rates Financial Crises Exchange Rates Healthcare Matters Litigation Matters Competition Matters Labor Disputes Intellectual Property Matters Fiscal Policy Taxes Government Spending, Deficits, and Debt Overall Average Entitlement and Welfare Programs Monetary Policy Regulation (Generic, Financial, Competition, Labor, Lawsuit) Generic Regulation (regulate, regulation, regulatory) Financial Regulation Competition Policy