What Triggers Stock Market Jumps?

Transcription

1 What Triggers Stock Market Jumps? Scott R. Baker (Kellogg, Northwestern) Nick Bloom (Stanford) Steven J. Davis (Chicago Booth) Marco Sammon (Kellogg, Northwestern) ASSA: January, 2019

2 1

3 Why does the stock market jump? Two broad views on what drives stock market moves: 1. Eugene Fama: driven by discount rates, dividends, etc. 2. Robert Shiller: Hard to explain fully by fundamentals; narratives develop and sometimes spread, affecting prices. 2

4 Previous Work on News Triggers of Stock Market Jumps Niederhoffer (JB, 1971) World news events drive jumps Cutler, Poterba and Summers (JPM, 1988) Many major news events with no associated jump 3

5 This Paper Analyzes Jumps on a Large Scale Study newspapers day after over /- 2.5% jumps across 15 countries back to 1980 (1900 for US) with 28 person RA team Main findings: Policy is important: 37% US jumps attributed to policy (and 26% internationally) US dominates globally: Outside US, newspapers attribute 34% of jumps to US above 20% US GDP share US share was especially high in the GFC Monetary Policy Jumps and Volatility: Realized volatility rises less after jumps triggered by monetary policy than after other jump categories Clarity Matters: Volatility is lower after jumps with clearer explanations 4

6 Outline Data: Triggers of Stock Market Jumps US Results: Stylized Facts International Results: Stylized Facts US Results: Implications of Different Jumps 5

7 Categorizing Jumps US jump definition: Days where the CRSP Value-Weighted Index has an absolute return of at least 2.5% Use the following day s newspaper to categorize jumps Later in sample, transition to same-day articles published online after market close To maximize accuracy Five papers read each day Each newspaper is read by at least 2 coders Created 136 page audit guide and RA training program Differences within and across papers discussed 6

8 Jump Categories Policy Categories Government Spending Taxes Monetary Policy & Central Banking Exchange Rate Policy & Capital Controls International Trade Policy Sovereign Military & Security Actions Regulation Elections & Political Transitions Other Policy Non-Policy Categories Macroeconomic News & Outlook Corporate Earnings & Outlook Commodities Foreign Stock Markets Unknown & No Explanation Terrorist Attacks & Large-Scale Violence by Non-State Actors Other Non-Policy No Article Found 7

9 Example (3/26/18, WSJ, 2.72%): Intl. Trade Policy This article would receive a primary category of International Trade Policy because the article links the rise to the reports of progress in the US-China trade talks. Geographic source would be the US and China. Journalist confidence would be High, as the article explicitly links the move to the trade talks. Ease of coding would be Easy. 8

10 Why Use Human Coders? Constructing training sample Economics Knowledge Language variation Topic Variation Nuances 9

11 How Reliable Are These Jump Codings? Cross-Coder and Cross-Newspaper Validation Policy vs. Non-Policy Granular Categories Observed Agreement Rate (All Coders & All Papers) Observed Agreement Rate (All Coders Within Paper) Agreement with Random Assignment 74% 42% 89% 71% 54% 14% 10

12 Additional Category-Level Validation Monetary Policy & Central Banking codings are more likely on FOMC announcement dates Macroeconomic News & Outlook codings are more likely on release dates for the Employment Situation Report and the CPI Report Elections & Political Transitions codings are more likely the day after national elections 11

13 Outline Data: Triggers of Stock Market Jumps US Results: Stylized Facts International Results: Stylized Facts US Results: Implications of Different Jumps 12

14 Category Breakdown for US: Category # Jumps % Jumps Macroeconomic News & Outlook % Unknown & No Explanation % Corporate Earnings & Outlook % Sovereign Military & Security Actions % Monetary Policy & Central Banking % Government Spending % Commodities % Regulation % Other Non-Policy % Elections & Political Transitions % Other Policy % Taxes % Exchange Rate Policy & Capital Controls % Foreign Stock Markets % International Trade Policy 8 0.7% Terrorist Attacks & Large-Scale Violence by Non-State Actors 8 0.7% No Article Found 5 0.5% 13

15 US Jumps by Year Depression Lowest Growth (1932) Panic 1901 Panic Banking Panic of 1907 WWI 1929 Crash Second Downturn (1937) WWII Oil Shock Black Monday Tech Boom/ Bust Global Financial Crisis Unknown + No Article Policy Non-Policy 14

16 Outline Data: Triggers of Stock Market Jumps US Results: Stylized Facts International Results: Stylized Facts US Results: Implications of Different Jumps 15

17 Category Breakdown for International Data: US Data: International Data: # Jumps % Jumps % Jumps Macroeconomic News & Outlook % 36.3% Unknown & No Explanation % 12.0% Corporate Earnings & Outlook % 14.5% Sovereign Military & Security Actions % 3.7% Monetary Policy & Central Banking % 12.7% Government Spending % 7.9% Commodities % 1.4% Regulation % 0.8% Other Non-Policy (Specify) % 2.8% Elections & Political Transitions % 1.8% Other Policy (Specify) % 1.5% Taxes % 1.3% Exchange Rate Policy, Capital Controls and International Trade Policy % 0.8% Foreign Stock Markets % 1.1% Terrorist Attacks & Large-Scale Violence by Non-State Actors % 1.5% No Article Found % 0.0% 16

18 Jump Shares Attributed to U.S. & Europe The sample for US Jump Share excludes the United States The sample for Europe Jump Share excludes European countries Early 1980 s Recession Black Monday Gulf War I Asia/LTCM Crises Tech Boom/Bust Global Financial Crisis European Debt Crisis Europe Jump Share Europe GDP Share US Jump Share US GDP Share 17

19 Outline Data: Triggers of Stock Market Jumps US Results: Stylized Facts International Results: Stylized Facts US Results: Implications of Different Jumps 18

20 Volatility is Higher After Macro Jumps and Lower After Monetary Policy Jumps, Relative to Other Jump Days Days After Jump Macro Monetary All Other 19

21 Going Beyond Categories Heavy selling was precipitated by the action of the Federal Reserve Bank of Boston in advancing its rediscount rate to 4% from 3.5%. WSJ, 11/10/1925, -3.7% There was nothing in the day s industrial and business news to account for the severe drop, and Wall street generally regarded it as due to an acute case of nerves. WSJ, 10/19/1929, -2.7% 20

22 4 Measures of Jump Clarity Pairwise Agreement Ease of Coding bandwidth = bandwidth =.2 Journalist Confidence Share Unknown bandwidth = bandwidth =.2 21

23 Average Clarity Has Been Increasing Over Time Clarity WWI Early 1900 s Panics 1929 Crash WWII Black Monday Tech Boom/ Bust Global Financial Crisis bandwidth =.2 22

24 Clear Moves Associated With Lower Volatility and Volume [Jump Day] Share of Total Total Distance Distance in Biggest 5- # of Within-Day Traveled Min. Window Cumulative Reversals Clarity ** *** * (0.0028) (0.0026) (0.0479) Observations R-Squared Elasticity Total distance traveled: Sum of 5 min. absolute returns Share: Biggest 5 min. return / total distance traveled Reversals: Cumulative return up to hour t+h different sign from cumulative return up to hour t+h+1. All specifications include standard finance controls 23

25 Volatility is Higher After Low Clarity Jumps Days After Jump Low Clarity High Clarity 24

26 2018 in Review Volatility Most annual jumps since 2011 Top 25% of years for number of jumps Clarity Top 15% in post-world War II era 2 of 9 jumps Unknown Lower confidence/ease of coding than post-wwii average Trade Policy 6 jumps attributed to trade policy between 1900 and jumps attributed to International Trade Policy in

27 Conclusion Main findings: Policy is important: 37% US jumps attributed to policy (and 26% internationally) US dominates globally: Outside US, newspapers attribute 34% of jumps to US above 20% US GDP share US share was especially high in the GFC Monetary Policy Jumps and Volatility: Realized volatility rises less after jumps triggered by monetary policy than after other jumps Clarity Matters: Volatility is lower after jumps with clearer explanations Next Steps: Analyze bond and exchange rate jumps, move threshold down to 2% for jumps in the US, extend international data Use jumps as an IV in macro-var to identify shocks Supervised machine learning analysis, using our human coders as a training sample 26

28 Back Up Slides 27

29 Policy Accounts for an Especially Large Share of Positive Jumps US: Share of Jumps Attributed to Policy Absolute Jump Size, Relative to Threshold Positive Negative Positive Negative + [0,0.5%) 43% 27% 36% 16% + [0.5%,1%) 43% 26% 46% 7% + [1%,1.5%) 40% 39% 46% 33% +2% or larger 53% 32% 53% 18% ROTW: Share of Jumps Attributed to Policy Absolute Jump (Overall) (of Known) Size, Relative to Threshold Positive Negative Positive Negative + [0,0.5%) 21% 22% 25% 25% + [0.5%,1%) 34% 16% 38% 17% + [1%,1.5%) 28% 22% 32% 25% +2% or larger 36% 23% 46% 27% 28

30 Additional Validation Exercises Industry-Level Jump Responsiveness: For many jumps, the explanation offered in next-day news accounts implies an amplified or muted response of certain industries to the news that moved the overall market. Example 1, Banks: During the GFC, the stock market responded positively to upward revisions in the likelihood or generosity of bank bailouts. For this type of jump, we expect an even more favorable response for Bank stocks. That is, the response of Banks is amplified relative to the overall market response. Example 2, Guns: When bad news about the likelihood or duration of the Iraq war triggered a negative jump, we expect the response for Guns (Defense firms) to be muted relative to overall market response. While a longer war may be bad for the overall U.S. economy, it is less bad (or even good) for Guns. Implementation: Use daily portfolio returns for 49 industries and consider all 339 US jumps from 1960 to News accounts of these jumps suggest an amplified or muted next-day response for 115 Jump-by-Industry returns. Results: We find strong, statistically significant evidence of amplified or muted industry-level return responses, as predicted. 29

31 Difference Between Monetary and Macro is Robust to Interaction with Positive/Negative Jumps Interaction with Jump Positive Interaction with Jump Negative Days After Jump Days After Jump Macro Monetary 30

32 Difference Between Macro and Monetary is Larger in Recessions.06 Interaction with Recession.04 Interaction with Not Recession Days After Jump Days After Jump Macro Monetary 31

33 Global Sample Selects Countries with Active Stock-Markets and Good On-Line Press Archives Country Start Sources Jump Threshold United States 1885 Wall Street Journal, etc. 2.50% United Kingdom 1930 Financial Times (UK Edition) 2.50% Australia 1985 Australian Financial Times 2.50% Canada 1980 The Globe and Mail 2.00% China (Hong Kong) 1988 South China Morning Post 3.80% China (Shanghai) 1994 Shanghai Securities Journal 4.00% Germany 1985 Handelsblat, FAZ 2.50% Greece 1989 Kathimerini, To Vima 4.00% Ireland 1987 The Irish Times 2.50% Japan 1981 Yomiuri and Asahi 3.00% New Zealand 1996 New Zealand Herald 2.50% Saudi Arabia 1994 Al Riyadh 2.50% Singapore 1980 Business Times and Straits Times 2.50% South Africa 1986 Business Day 2.50% South Korea 1980 Chosun Ilbo 2.50% Jump threshold was chosen such that jumps were approximately 1% of trading days 32

34 Count by Year for the UK (Policy 33%) Great Depression WWII Suez Crisis Recession and 1976 IMF crisis Sterling Crisis Black Monday Tech boom/ bust Global Financial Crisis Unknown + No Article Policy Non-Policy Notes: Each bar represents the number of jumps from each category within a given year 33

35 Validation Detail, 1 1. Industry-Level Jump Responsiveness: For many jumps, the explanation offered in next-day accounts implies an amplified or dampened response of certain industries to the news that moved the overall market. Example 1, Banks: During the GFC, the stock market responded positively to upward revisions in the likelihood or generosity of bank bailouts. For this type of jump, we expect an even more favorable response for Bank stocks. That is, the response of Banks is AMPLIFIED relative to the overall market response. Example 2, Guns: When bad news about the likelihood or duration of the Iraq war generated a negative jump, we expect the response for Guns (Defense firms) to be DAMPENED relative to the overall market response. While a longer war may be bad for the overall U.S. economy, it is less bad (or even good) for Guns.

36 Validation Detail, 2 Implementation: Obtain daily portfolio returns for 49 industries. Review coding detail for U.S. equity market jumps from 1960 to If the detailed description for jump-date t implicates industry i, then assign values to!"# $% as follows:!"# $% = 1, if jump description implies AMPLIFIED response of i. = -1, if jump description implies DAMPENED response of I; = 0, OTHERWISE. In making these industry assignments, we take a conservative approach, as follows: We typically make industry assignments based on the Primary jump reason only, not the Secondary Jump reason (if there is a Secondary reason. We set!"# to 0 when the detailed explanation for the Jump involved an overly broad industry group for our purposes. For example, Manufacturing, maps to at least 15 of the 49 industry groups.

37 Validation Detail, 3 Most jumps do not map readily to a particular industry. Sometimes, we assign 2 industries to a given jump. Most, but not all, of these dual assignments involve Sovereign Military Jumps, which implicate both Guns and Aerospace. Among our 339 jumps from 1960 to 2016, we obtain 115 Jump-Day X Industry observations with nonzero Tri values, as follows: Banks: 38 nonzero values Guns: 19 Aerospace: 16 Others, all with less than 10 nonzero Tri values: Oil, Coal, Building Materials, Construction, Autos, Chips, Hardware, Household Goods, Software, Electrical Equipment " #$ = the daily return for industry portfolio i on day t. &" $ = the daily return on market portfolio on day t.

38 Validation Detail, 4 One-industry-at-a-time approach Consider daily industry-level returns for i : 8 9: = < + >?8 : 9: + DABC 9:?8 : + E : Pooled-sample approach Consider daily industry-level returns for industries with at least 3 nonzero Tri values: 8 9: = 9 < > 9?8 : + 9 ABC 9: + DABC 9:?8 : + E : Under both approaches, the hypothesis of interest is L M : D = 0, L P : D > 0

39 Validation Detail, 5 Banks Pooled Sample All Days Jump Days All Days Jump Days! Coefficient 0.80*** 0.74*** 0.55*** 0.51*** (St. Error) (0.23) (0.24) (0.13) (0.13) Observations 13, , R-Squared A regression for Guns, yields results similar to the Pooled ones, but the standard error is large and the coefficient estimate is insignificant. When we set Tri=-1 for the Aerospace industry for jumps attributed to Sovereign Military Conflict, the Aerospace regression yields a small, marginally significant coefficient of the wrong sign. That may reflect the ambiguous nature of Aerospace firms responses to military conflict: (relatively) good news for defense-oriented aerospace firms may, at the same time, be bad for aerospace firms oriented toward civilian customers. If we set Tri=1 for Aerospace in these cases, the anomalous Aerospace result disappears, and the Pooled Sample results get stronger.

40 Validation Detail, 6 The foregoing results confirm that our classification of jumps (based on next-day newspaper accounts) reflects useful information about what actually triggered the jump. 2.Announcement-Date Responses: On the next slide, we test whether jumps attributed to: Monetary Policy & Central Banking are more likely on FOMC announcement dates (Yes) Macroeconomic News & Outlook are more likely on release dates for the Employment Situation Report and the CPI Report (Yes) Elections & Political Transitions are more likely the day after national elections (Yes)

41 Validation Detail, 7 Dependent Variable: Indicated Jump Coding x 100 (1) Monetary Policy (2) Macro (3) Elections # Known Dates FOMC Meeting Date 1.39** 159 or Next Day (0.693) FOMC Press Release 3.65*** Date (1.309) (0.590) (0.277) CPI or Employment ** -0.10*** 827 Situation Release Date (0.499) (0.297) (0.455) (0.020) Day After National *** ** 49 Elections (1.038) (0.241) (2.246) (1.992) Constant 0.19*** 0.34*** 0.83*** 0.09*** (0.072) (0.073) (0.076) (0.017) # Codings Observations 3,288 5,792 12,838 22,929 R-Squared Notes: Each column (1) to (3) reports a regression of jump coding values (times 100) for the indicated category on a set of known information-release dates. The results show that our newspaper-based attributions of jumps to (1) Monetary Policy, (2) Macro News & Outlook, and (3) Elections & Political Transitions occur with greater relative and absolute frequency on FOMC Press Release Dates, CPI or Employment Situation Release Dates, and the Day After National Elections, respectively. Results are similar when adding day-of-week controls.

42 Monetary Policy Category More Likely Following FOMC Meetings Baseline + HAR Interaction ` All Days Jump Days All Days Jump Days Positive Jumps Negative Jumps FOMC Meeting at t or t-1 (FOMC_t) 2.13** 33.34*** *** 43.62*** (0.84) (10.99) (0.99) (12.10) (14.99) (34.17) Avg. return over past 6 weeks *** *** (81.62) (834.40) (79.24) (835.45) ( ) (784.53) Avg. return over past 6 weeks * FOMC_t ,765.68* (610.23) ( ) (614.41) ( ) ( ) ( ) Vol. over last six weeks *** *** ** (33.64) (112.87) (33.15) (112.69) (214.58) (86.72) Vol. over last six weeks * FOMC_t (209.01) (648.83) (726.37) ( ) Constant 0.25** 20.69*** 0.30** 20.79*** 26.24*** 15.71*** (0.13) (3.19) (0.12) (3.21) (5.37) (3.81) Observations 9, , R-squared Notes: LHS is 100 times an indicator variable for a Monetary Policy Jump. US data, 1981 to present. FOMC_t only includes scheduled meetings. 41

43 Foreign stock markets News reports that attribute a large domestic market move directly to foreign stock-market moves without offering any deeper explanation for the domestic or foreign stock market move. For example, suppose an article in the South China Morning Post attributes a jump in the Hang Seng (Hong Kong) Index to stock price moves on Wall Street. If the article offers no explanation for the Wall Street move and no other explanation for the Hang Seng jump, code the Primary Category as Foreign Stock Markets. If instead, the article attributes the Wall Street moves to bad reports about the U.S. macroeconomic outlook, for example, code the Primary Category as Macroeconomic News & Outlook. 42

44 Foreign Stock Markets 1 Turnover crosses $10b mark as players plough into battered sector Buyers take cue from US Wilder, David. South China Morning Post; Hong Kong [Hong Kong]12 Apr 2001: 1. Hong Kong technology stocks led the market higher yesterday after a sharp bounce on Wall Street triggered a buying spree in the beleaguered sector. The Hang Seng Index jumped 4.03 per cent to end at 12, points, with turnover at a relatively sprightly HK$10.14 billion - its first foray above the $10 billion mark since the March 23 trading session. "We hope that there will be more rebounds in the [United States] and then this market should move further," BNP Prime Peregrine sales trader Hugo Leung said. "Looking at the turnover, it's definitely improved." On Tuesday, the Nasdaq Stock Market soared 6.09 per cent and the Dow Jones Industrial Average climbed 2.61 per cent to end back above the 10,000-point level. This in turn sparked a rebound in many Hong Kong stocks and, unlike other recent bounces, small-caps came along for the ride. Systems integrator Computer and Technologies, for instance, soared per cent to $ The Primary Category is Foreign Stock Markets, because the article attributes the jump in the Hang Seng Index to rising U.S. stock prices, and the full article says nothing about any deeper factor behind the stock market rise in the HK or the U.S. Journalist Confidence is High, given the even stronger attribution in the second highlighted passage. Geographic Origin is USA. 43

45 Foreign Stock Markets 2 The Primary Category is Foreign Stock Markets, because the article describes the market movements in Shanghai as the catalyst for the U.S. equity market drop, and it offers no deeper explanation for the drop in the Shanghai or US markets. Journalist Confidence is High, because it clearly describes Shanghai s markets as being 44 responsible, and refers to this reasoning later in the article as well.

46 Average Within Paper Agreement Granular Categories Policy vs. Non-Policy Granular Categories:.71, Policy vs. Non-Policy:.89 45

47 Average Pairwise Agreement by Year All Papers Granular Categories Policy vs. Non-Policy Granular Categories:.42, Policy vs. Non-Policy:.74 Randomly drawing jump categories from unconditional distribution, agreement is 54% on policy vs. non-policy and 13.5% on granular categories 46

48 Triggers of Losses by Era Years # of jumps Most Common 2nd Most Pre-Fed Era Unknown Corp. Earnings World War I Sov. Military Macro. News 1920s Unknown Macro. News Depression Era Macro. News Unknown World War II Sov. Military Macro. News Early Postwar Macro. News Sov. Military Inflation & Oil shocks Macro. News Commodities Disinflation & Growth Macro. News Corp. Earnings Boom, Rec. & Recovery Macro. News Corp. Earnings Global Financial Crisis Macro. News Corp. Earnings Post GFC Macro. News Unknown All Periods Macro. News Unknown Notes: We identify the all negative stock market jumps in each era, and identify the modal categories among these moves. 47

49 Triggers of Gains by Era Years # of jumps Most Common 2nd Most Pre-Fed Era Unknown Corp. Earnings World War I Sov. Military Unknown 1920s Unknown Corp. Earnings Depression Era Macro. News Monetary Policy World War II Sov. Military Monetary Policy Early Postwar Unknown Sov. Military Inflation & Oil shocks Monetary Policy Macro. News Disinflation & Growth Macro. News Monetary Policy Boom, Rec. & Recovery Monetary Policy Macro. News Global Financial Crisis Macro. News Corp. Earnings Post GFC Macro. News Monetary Policy All Periods Macro. News Unknown Notes: We identify all positive stock market jumps in each era and identify the modal categories among these moves. 48

50 Triggers of Biggest Gains/Losses by Era Negative Jumps Positive Jumps # jumps Years Most Common 2nd Most Most Common 2nd Most Pre-Fed Era Unknown Corp. Earnings Unknown Corp. Earnings World War I Sov. Military Macro. News Sov. Military Unknown 1920s Unknown Macro. News Unknown Corp. Earnings Depression Era Macro. News Unknown Macro. News Monetary Policy World War II Sov. Military Macro. News Sov. Military Monetary Policy Early Postwar Macro. News Sov. Military Unknown Sov. Military Inflation & Oil shocks Macro. News Commodities Monetary Policy Macro. News Disinflation & Growth Macro. News Corp. Earnings Macro. News Monetary Policy Boom, Rec. & Recovery Macro. News Corp. Earnings Monetary Policy Macro. News Global Financial Crisis Macro. News Corp. Earnings Macro. News Corp. Earnings Post GFC Macro. News Unknown Macro. News Monetary Policy All Periods Macro. News Unknown Macro. News Unknown Notes: We identify the 10 biggest stock market gains/losses in each era, and identify the modal categories among these moves. 49

51 Macro Jumps Predict Future Real Outcomes Response: GDP Impulse: # Negative Macro Jumps Impulse: # Positive Macro Jumps Quarters Quarters US Data, VAR with GDP, # Positive and # Negative Macro Jumps. 50

52 For Non-Macro Jumps, Volatility is King Response: GDP 100 Impulse: # Negative Non-Macro Jumps 50 Impulse: # Positive Non-Macro Jumps Quarters Quarters US Data, VAR with GDP, # Positive and # Negative Non-Macro Jumps. 51