Difference between revisions of "2020 Stock Market Crash"
| Line 1: | Line 1: | ||
{{Project | {{Project | ||
|Has title=2020 Stock Market Crash | |Has title=2020 Stock Market Crash | ||
| + | |Has image=Trader-city-london-trading-stock-5129770.jpg | ||
|Has owner=Ed Egan | |Has owner=Ed Egan | ||
|Has project output=Content | |Has project output=Content | ||
Revision as of 11:45, 13 October 2020
| 2020 Stock Market Crash | |
|---|---|
| Project Information | |
| Has title | 2020 Stock Market Crash |
| Has owner | Ed Egan |
| Has start date | |
| Has deadline date | |
| Has project status | Complete |
| Has sponsor | edegan.com |
| File location(s): | E:\projects\stockmarketanalysis |
| Has project output | Content |
| Copyright © 2019 edegan.com. All Rights Reserved. | |
This page details my exploration of data concerning the 2020 Stock Market Crash. This work spawned a short-form academic paper, Empirical Regularities in Stock Market Crashes, as well as a number of Op. Ed. submissions. I ultimately put a lay-version of the historical crash analysis up on my blog.
Summary
Between the 19th of February, 2020, and the 23rd of March, 2020, the S&P500 (^GSPC) fell from 3,386.15 to 2,237.40, using adjusted closing prices. It subsequently climbed back to 3,276.02 on the 22nd of July, 2020. Some people seem to think it will keep going. I expect a massive retrenchment.
To predict the characteristics of the 2020 stock market crash, I propose examining the market using three techniques:
- Historical crash analysis
- Price-to-earnings (or revenue, etc.) analysis that attempts to factor defaults
- Extrapolating based on the market-to-economy relationship
I could also employ a pure chartist approach, perhaps to give the counterfactual.
Historical Crash Analysis
Scripts and Analysis
The SQL script Analysis.sql is in E:\projects\stockmarket, as are the source files.
The dbase is stockmarket, and is loaded on mother. It's source files are in Bulk\stockmarket.
It loads the DJIA, the GSPC, and the crashes file. It will also load the COMPUSTAT pull.
Op Ed Submissions
I submitted the following:
- Wall Street Journal, Aug 8th. Rejection by silence: Egan (2020) - Category 3 Stock Market Crash (WSJ Op Ed V2-1).docx
- New York Times, Aug 13th. Rejection by silence: Egan (2020) - Category 3 Stock Market Crash (WSJ Op Ed V2-2).docx
- Finacial Times, Aug 28th, Rejection by silence: Egan (2020) - What to Learn from Past Stock Market Crashes (FT V3).docx
- Washington Post, Sept 1st, Rejected by email: Egan (2020) - What to Learn from Past Stock Market Crashes (WP V2).docx
References the Op Eds
- Forbes: https://www.forbes.com/sites/sergeiklebnikov/2020/07/23/historical-stock-market-returns-under-every-us-president/#e5c3268faaf4
- WSJ: https://www.wsj.com/articles/stock-scenarios-for-a-madcap-year-11597607961
- Business insider: https://www.businessinsider.com/stock-market-coronavirus-crash-not-over-how-to-invest-bofa-2020-3
Checking results
I wrote an Op-Ed and submitted it to the WSJ on August 8th. It is in the project folder:
E:\projects\stockmarketanalysis\Egan (2020) - Category 3 Stock Market Crash (WSJ Op Ed V2-2).docx (and .pdf)
In that article, I included a table of durations, multiples lost, and max declines. I want to check the durations...
- First check: Did I double load the index file? Probably not as some durations are odd. DJIAFull.txt has 31127 lines. Table djia has 31126 record, all distinct.
- Second check: Durations are calculated by counting records WHERE WHERE reltrading >ppreltrading AND reltrading < refindreltradingfirst in DJIACrashFirst, etc. These queries come from DJIACrashWMarkers, which is blown out by Event (and has 85,732 records). I checked the Asian Contagion and Black Monday. They were fine.
- Third check: I manually checked the index file for and calculated the duration of the dot com crash:
- I found a -6% on April 14th,2000. The index was a 10305. I worked back.
- The prior peak was on Jan 17th, 2000. It was 11722. I worked forward.
- The next time the index hit that level was Mar 10th, 2006. There were 1,699 trading days in the interim, which is what I have in the report.
- It then stayed above that level until June 26th, 2008... So one crash blends into the next.
To Do:
- Calculate average annual growth rate and compare it to max decline
- Calculate tdays before and after max decline per crash to measure asymmetry.
For some reason, the price is missing for 10/26/1940. This causes a #DIV/0! error in excel. I removed it. The average daily return is then 0.0275% and the average annual return (using 252 trading days) is 6.94%. From 1/2/1985 forward, this average is 0.0411% per day or 10.35% per year.
It seems that, on average, a third (actually 31%) of a crash occurs before the maximum decline and 2/3rds (actually 69%) occurs afterwards, however, the location of the maximum decline looks pretty random.
Data
Index Data
As much as I love the Dow, it accounts for about 25% of the economy, whereas the S&P500 accounts for around 80%. I'll download both from Yahoo:
- https://finance.yahoo.com/quote/%5EGSPC/history?period1=157766400&period2=1595548800&interval=1d&filter=history&frequency=1d
- https://finance.yahoo.com/quote/%5EDJI/history?period1=157766400&period2=1595548800&interval=1d&filter=history&frequency=1d
Trying for older data is problematic from Yahoo. The site complains about start dates being ahead of end dates with start dates before 1976ish... Usually, the issue doesn't happen with 2digit date less than 41. You can edit the URL directly using Unix Timestamps to create 1900 to 28th July 2020: https://finance.yahoo.com/quote/%5EDJI/history?period1=-2208988800&period2=1595980800&interval=1d&filter=history&frequency=1d. However, the data you get will still start on 1/29/1985. The Dow started in May 26, 1896 [1]. The WSJ gives data from 1970 forward... However, historical data is available from WRDS for the DJI from May 26th 1896 to Dec 31st 2007. I downloaded it as dbe3d1bd4ba151a3.txt (525 KB, 27955 observations 3 variables).
S&P500 identifiers have apparently become proprietary - WRDS no longer carries them. Also the S&P500 started in 1957, which is long after the Wall Street Crash (I'd like to include this crash!), and historical data is now hard to find. Overall, I think I'll go back the Dow!
Interest Rate Spread
The St. Louis Fed has data on Aaa and Baa Corporate Bonds:
Lit Review
It seems that there's a non-econ based literature that tries to predict crashes using various statistical techniques. A bunch of these papers appear in the journal "Physica A: Statistical Mechanics and its Applications" (WX Zhou abd D Sornette authored 3 of them, and Sornette wrote a book called "Why stock markets crash: critical events in complex financial systems" [2]). See, for example:
- Can one make any crash prediction in finance using the local Hurst exponent idea? [3] *Antibubble and prediction of China's stock market and real-estate [4]
- Can we predict crashes? The case of the Brazilian stock market [5]
- Predicting critical crashes? A new restriction for the free variables [6]
- Renormalization group analysis of the 2000–2002 anti-bubble in the US S&P500 index: explanation of the hierarchy of five crashes and prediction [7]
- Autoregressive conditional duration as a model for financial market crashes prediction
- The use of the Hurst exponent to predict changes in trends on the Warsaw Stock Exchange
- Stock market dynamics: Before and after stock market crashes
Others in non-finance/econ journals include:
- The Nasdaq crash of April 2000: Yet another example of log-periodicity in a speculative bubble ending in a crash [8]
- Use of machine learning algorithms and twitter sentiment analysis for stock market prediction [9]
- [PDF] Application of neural networks to stock market prediction, AS Kulkarni - Technical Report, 1996 - machine-learning.martinsewell.com [10]
- Forecasting stock market crisis events using deep and statistical machine learning techniques, SP Chatzis, V Siakoulis, A Petropoulos… - Expert systems with …, 2018 - Elsevier
- [BOOK] Stock Market Crashes: Predictable and Unpredictable and What to Do About Them [11]
In finance, econ, and business journals:
- The “CAPS” prediction system and stock market returns, CN Avery, JA Chevalier, RJ Zeckhauser - Review of Finance, 2016 [12]
- Testing for financial crashes using the log periodic power law model DS Brée, NL Joseph - International review of financial analysis, 2013 - Elsevier
- Naive trading rules in financial markets and wiener-kolmogorov prediction theory: a study of" technical analysis" SN Neftci - Journal of Business, 1991 - JSTOR [13]
- Stock market bubbles in the laboratory, DP Porter, VL Smith - Applied mathematical finance, 1994 - Taylor & Francis
- Investor behavior in the October 1987 stock market crash: Survey evidence, RJ Shiller - 1987 - nber.org [14]
- Bubble diagnosis and prediction of the 2005–2007 and 2008–2009 Chinese stock market bubbles, ZQ Jiang, WX Zhou, D Sornette, R Woodard… - Journal of economic …, 2010 - Elsevier [15]
- Stock market crashes in 2007–2009: were we able to predict them?, S Lleo, WT Ziemba - Quantitative Finance, 2012 - Taylor & Francis
- Market volatility prediction and the efficiency of the S & P 100 index option market, CR Harvey, RE Whaley - Journal of Financial Economics, 1992 - Elsevier [16]
- Does trading volume contain information to predict stock returns? Evidence from China's stock markets, CF Lee, OM Rui - Review of Quantitative Finance and Accounting, 2000 - Springer [17]
- US stock market crash risk, 1926–2010, DS Bates - Journal of Financial Economics, 2012 - Elsevier
- Triggering the 1987 stock market crash: Antitakeover provisions in the proposed house ways and means tax bill? ML Mitchell, JM Netter - Journal of Financial Economics, 1989 - Elsevier [18]
- To what extent did stock index futures contribute to the October 1987 stock market crash, A Antoniou, I Garrett - The Economic Journal, 1993 - academic.oup.com
- [PDF] A mean-reversion theory of stock-market crashes, E Hillebrand - Journal of Finance, 2003 - cofar.uni-mainz.de [19]
- Explaining what leads up to stock market crashes: A phase transition model and scalability dynamics, R Yalamova, B McKelvey - Journal of Behavioral Finance, 2011 - Taylor & Francis
- Speculative bubbles, crashes and rational expectations, OJ Blanchard - Economics letters, 1979 - Elsevier
Particularly useful papers might include:
- US stock market crashes and their aftermath: implications for monetary policy, FS Mishkin, EN White - 2002 - nber.org [20]
- This paper examines fifteen historical episodes of stock market crashes and their aftermath in the United States over the last one hundred years.
- Stock-market crashes and depressions, RJ Barro, JF Ursúa - 2009 - nber.org [21]
- Long-term data for 30 countries up to 2006 reveal 232 stock-market crashes (multi-year real returns of–25% or less) and 100 depressions (multi-year macroeconomic declines of 10% or more), with 71 of the cases matched by timing. The United States has two of the matched …
- [BOOK] Stock market crashes and speculative manias, EN White - 1996 - econpapers.repec.org [22]
- This volume offers an authoritiative selection of the best published articles on the great speculative manias and stock market crashes, which highlights their important similarities. These phenomena disrupt the normal activities of investors who use financial markets to …
- The predictability of stock market regime: evidence from the Toronto Stock Exchange,S Van Norden, H Schaller - The Review of Economics and Statistics, 1993 - JSTOR
- Thus the existence of bubbles would not only account for occasional asset price crashes but also rapid run-ups in asset prices before a crash. The question we address is whether or not stock market crashes and the booms that precede them are related to apparent deviations …
- Stock market crashes and dynamics of aftershocks, P Kapopoulos, F Siokis - Economics Letters, 2005 - Elsevier [23]
- We begin with the intuitive observation that short-term business-as-usual process and bubble rising looks like an accelerated energy before an earthquake. In such a framework, the aftershocks resemble the correction process of the stock market. We investigate the statistical properties of stock returns in the financial markets just after a major market crash. It is found that the aftershocks obey the well-known Gutenberg–Richter simple rule in geophysics. Our empirical observations show that the statistical properties of aftershocks sequences in the crashes of late '90s and early '00s are essentially different from the ones observed a decade earlier.
Questions and Answers
- Is there an objective definition of a crash?
- Barro and Ursua 2009: multi-year real returns of –25% or less. They also define a depression as multi-year macroeconomic declines of 10% or more. Crashes and correlated with minor depressions 31% of the time, and major depressions 10% of the time. In reverse, minor depressions are associated with crashes 71% of the time and major ones 92% of the time.... Part of this analysis entails splitting the sample into different “bins” or categories, defined by the magnitudes of the stock market crashes or by the type of environment that characterized those events; for example, wars, currency or banking crises, and periods of global economic distress.
- Mishkin and White 2002: On the face of it, defining a stock market crash or collapse is simple. When you see it, you know it. However, attempting a more precise definition and measurement over the course of a century is more difficult. ... As both fell slightly over 20 percent, a 20 percent drop in the market is used to define a stock market crash. The fall in the market, the depth is, however, only one characteristic of a crash. Speed is another feature. Therefore, we look at declines over windows of one day, five days, one month, three months, and one year...
- Is there a taxonomy of stock market crashes?
- Mishkin and White 2002: The description of the fifteen episodes of twentieth-century stock market crashes suggests that we can place them into four categories. 1. Episodes in which the crashes did not appear to put stress on the financial system because interest-rate spreads did not widen appreciably. These include the crashes of 1903, 1940, 1946, 1962 and 2000. 2. Episodes in which the crashes were extremely sharp and which put stress on the 31 financial system, but where there was little widening of spreads subsequently because of intervention by the Federal Reserve to keep the financial system functioning in the wake of these crashes. These include the crashes of 1929 and 1987. 3. Episodes in which the crashes were associated with large increases in spreads suggesting severe financial distress. These include the crashes of 1907, 1930-33, 1937, 1973-74. 4. Episodes in which the crashes were associated with increases in spreads that were not as large as in the third category, suggesting some financial distress. These include the crashes of 1917, 1920, 1969-70 and 1990.
- Are there standard measures of a crash?
- First drop from peak
- Largest drop from peak
- Time to first recovery to peak
- Time to sustained recovery to peak
- Off-performance area (i.e., days x difference from the peak?) Should I factor in expected growth?
- Number of drops? Precursory volatility? Total volatility?
Scripts to build:
- Identify a slump:
- Automated: could calculate all 1 day, 1 week, 1 month, 3 month, or 1-year cumulative drops and look for those of more than 20% a la Mishkin vs. 25% a la using Barro.
- Find the previous peak.
- Find any transitive returns to the peak.
- Find the 'permanent' return to peak (it turns out that you can go from crash to crash).
- Calculate area lost in point-days and percent-days.
Other Notes
Exploring Complexity
Self-Organizing Criticality, which gives rise to Scale Invariance, might provide a model. See the Abelian sandpile model as a place to start.
Playing with Crashes
This is a legacy subsection. It was superseded by the approach embodied in the scripts. I've left it here because it provides some useful base data for readers interested in the problem.
Data before the 1980s probably isn't useful. It was a different world back then, both in terms of globalization and in terms of how stock markets worked. So, I'll start with the 1987 crash and go forward from there. I originally tried to suggest three classes of crash, like hurricanes:
| Class | Description |
|---|---|
| 1 | Flash crashes and other short-term market anomalies |
| 2 | Short to medium term economic disturbances that are limited to a sector |
| 3 | Economic crises |
My starting list comes from a list of stock market crashes.
Using the largest decline on the Dow of the year (except for 1994) as the crash day gives: Result(Copy to clipboard)
| Event | Start Year | Class | Largest SingleDay | Date | Notes |
|---|---|---|---|---|---|
| Black Monday | 1987 | 3 | -22.60% | 10/19/1987 | A -3.8% in 3 days and an -8% follows a week later. Preceeded by a -4.6% 3 days earlier, and a -3.8% 2 days before that. |
| Early 90s recession | 1990 | 2 | -3.30% | 8/6/1990 | It plummets even more on Nov 15th of 1991, with a 3.8% |
| Part II of the early 90s | 1994 | 2 | -2.40% | 2/4/1994 | Almost the same drop on Nov 22nd (actually slightly larger), later in the year |
| Dot Com Crash | 2000 | 3 | -5.70% | 4/14/2000 | A -3.7% five weeks earlier |
| Financial crisis | 2007 | 2 | -3.30% | 2/27/2007 | Lot of volatility later in the year |
| Financial crisis II | 2008 | 3 | -7.90% | 10/15/2008 | Series of ~4% drops in Sept and then some 5-6% ones in Oct/Nov. |
| 2010 Flash Crash | 2010 | 1 | -3.60% | 5/20/2010 | Were -3.2% on 6th May and on 4th June |
| Aug 2011 Fall | 2011 | 1 | -5.50% | 8/8/2011 | First drop was on 4th Aug (-4.3%), then 8th, then 10th (-4.6%). Other bid declines in Sept and Nov. |
| 2015–16 stock market selloff | 2015 | 1 | -3.60% | 8/24/2015 | Was a -3.1% 3 days earlier and a -2.8% a week later |
| Asian Contagion | 1997 | 1 | -7.2% | 10/27/1997 | Had been a 3.1% decline in Aug |
| 1998 Bear Market | 1998 | 2 | -6.40% | 8/31/1998 | Was a -4.2% 4 days earlier, a a -3.4% on the 4th before that. It was also followed by a -3.2% on the 10th Sept. |
| Friday 13th Mini-crash | 1989 | 1 | -6.90% | 10/13/1989 | Appears isolated |
Data driven approach
Identify crashes as using the largest single-day declines on the DJIA:
| Date | Adj Close | Volume | SingleDayChange |
|---|---|---|---|
| 10/19/1987 | 1738.73999 | 87230000 | -22.6% |
| 10/26/1987 | 1793.930054 | 35420000 | -8.0% |
| 1/8/1988 | 1911.310059 | 27440000 | -6.9% |
| 10/13/1989 | 2569.26001 | 37620000 | -6.9% |
| 10/27/1997 | 7161.200195 | 91830000 | -7.2% |
| 8/31/1998 | 7539.069824 | 117890000 | -6.4% |
| 4/14/2000 | 10305.76953 | 267580000 | -5.7% |
| 9/17/2001 | 8920.700195 | 565600000 | -7.1% |
| 9/29/2008 | 10365.4502 | 385940000 | -7.0% |
| 10/7/2008 | 9447.110352 | 362520000 | -5.1% |
| 10/9/2008 | 8579.19043 | 436740000 | -7.3% |
| 10/15/2008 | 8577.910156 | 374350000 | -7.9% |
| 10/22/2008 | 8519.209961 | 348840000 | -5.7% |
| 11/5/2008 | 9139.269531 | 264640000 | -5.0% |
| 11/19/2008 | 7997.279785 | 350470000 | -5.1% |
| 11/20/2008 | 7552.290039 | 528130000 | -5.6% |
| 12/1/2008 | 8149.089844 | 321010000 | -7.7% |
| 8/8/2011 | 10809.84961 | 479980000 | -5.5% |
| 3/9/2020 | 23851.01953 | 750430000 | -7.8% |
| 3/11/2020 | 23553.2207 | 663960000 | -5.9% |
| 3/12/2020 | 21200.61914 | 908260000 | -10.0% |
| 3/16/2020 | 20188.51953 | 770130000 | -12.9% |
| 3/18/2020 | 19898.91992 | 871360000 | -6.3% |
| 6/11/2020 | 25128.16992 | 647780000 | -6.9% |
Grouping them using 6-month windows gives 9 crashes:
- 1987/1988: 3 obs (including largest)
- 1989: 1 obs
- 1997: 1 obs
- 1998: 1 obs
- 2000: 1 obs
- 2001: 1 obs
- 2008: 9 obs
- 2011: 1 obs
- 2020: 6 obs (and counting, including second and third largest)
I revamped this approach to find 12 crashes (14 including the current one and the 1929 Wall Street Crash) and the data naturally groups itself into 4 categories. See the scripts.
PE Analysis
COMPUSTAT data
COMPUSTAT updates its fundamentals quarterly and monthly security files daily. I did a pull of the entire fundamentals quarterly database from 2015-01 to 2020-07. It has close prices, as well as highs and lows. I pulled close, as averaging highs and lows seems like it might be skewed given the context (especially over short time periods).
