How To Predict Market Capitulation Days (And How To Profit)

Tyler Durden's picture

The Goldman Markets team has put together this handy analysis on a proposed set of metrics that could be used to spot capitulation days, based on such inputs as VIX, options skew, daily trading volume relative to the 50 DMA, magnitude of sell-off,
return from S&P’s intraday low to close, and the Fama-French winner/loser momentum. Based on the correct spotting of true capitulation days, Goldman predicts that those buying stocks following dramatic sell offs leads to abnormal profits. On the other hand Goldman refuses to mention the alternative: shorting the market on days when stocks, which now have an implied correlation of about 1 and no associated volume, melt up alongside the risk-FX squeeze. Furthermore, there is no accounting of those situations where after a several hundred point drop, the Federal Reserve refuses to get involved, which as Alan Greenspan pointed out, is unlikely, due to the Fed's perception that the market is the one true indicator of economic health (and nothing less than a case of the tail wagging the dog.) Lastly, one should not forget the abysmal track record of Markets' group top 2010 trade recommendations, presented below as of today: with 2 out of 9 top recommendations profitable, clients who have bet against Goldman's sell side advice have made a mint.

Can Capitulation Days Be Predicted?  A Follow-Up

From time to time, we also look at interesting statistical methods to complement our fundamental work.  With investors reading the tea leaves for signs of a market bottom back in early July, we revisit our earlier work from July 2008 that distinguishes between the “true capitulation days” – days that mark the end of downward trend in equities – and days that are just stepping stones to further losses.

Using S&P 500 data back to 1990, the screening process consisted of first isolating the “candidate capitulation days”, which we defined as days when S&P traded more than 5% below its 50-day MA, 5% below prior week’s close and 1% below previous day’s close.  Then, to qualify as a “true capitulation day”, the “candidate capitulation day” had to turn positive the following day, gain at least 1% the following week, and gain at least 3% the following month.  In the end, the above described process filtered out 188 “candidate capitulation days” and 37 “true capitulation days” for us.  The criteria used here to screen for “true capitation days” is obviously arbitrary, but our sensitivity analysis showed little impact on our derived results from varying the parameters.

Secondly, to improve the odds of ex-ante indentifying a “true capitulation day”, we consider six additional factors that may influence the probability of correctly detecting a “true capitulation day”.  Those factors are: 1) the VIX, 2) options skew, 3) daily trading volume relative to its 50d MA, 4) magnitude of sell-off, 5) return from S&P’s intraday low to close, and finally 6) the so-called Fama-French winner/loser momentum.  Thus, conditional on observing a ‘candidate’ capitulation day, these factors were useful in helping us identify whether an event may turn out to be ‘true’, i.e., the market does better thereafter.

In our earlier work, we concluded that the VIX and cross-sectional momentum gave the clearest signals, while trading volume and options skew gave the most mixed results.  But the 2008 experience countered some of those lessons, with the VIX peaking in November ’08, while S&P only bottomed in March of ’09.

New Systematic Approach to Predicting Capitulation Days

To try and create a systematic screening tool for “true capitulation days”, we estimate a simple econometric model using five factors to come up with the likelihood of identifying a “true capitulation day”.  Specifically, we use: 1) the VIX, 2) daily trading volume relative to its 50d MA, 3) return from S&P’s intraday low to close, 4) winner/loser momentum and 5) VIX 2-months - 1-month  time spreads to filter out days, whose characteristics most closely match those of “true capitulation days.”   The goodness of fit of this simple “probit model” is modest (on the order of 30%).  But all the coefficients matter (i.e. they are statistically significant).  The resulting output of this tool is a daily probability from 0% to 100%, where 100% suggests that that a day resembles a “true capitulation day” the most, while 0% suggests it resembles a “true capitulation day” the least.

It Pays To Pay Attention to Market Capitulation: Real-Time Perception vs. Ex-Post Reality

A preliminary look at the predictive power of our tool on subsequent S&P 500 returns provides promising results, which deserve to be studied in more detail along with their significance relative to prevailing volatility in the markets.  We compared the returns across quartiles of our “capitulation probability” metric.  Indeed, the subsequent returns tend to diminish as we move from days marked as “most resembling true capitulation days” to “least” across the spectrum of our probabilities.  In fact, the average 1, 2 & 3-months subsequent returns for the top quartile days (capitulation day probability of at least 75%) are 1.2%, 1.8% and 2.4%.  On the other hand, the average returns for the bottom quartile (capitulation day probability of at most 25%) for the same time horizons are 0%, 0.3% and 0.6%.

A better test of whether these factors capture the market's mood is whether a meaningful concentration of days that closely resemble “true capitulation days” in one month tends to drive the conviction that the market has capitulated?

We simulate this theory using our model by designating days with “capitulation probability” of at least 75% as those with high perception of market capitulation and evenly dividing our sample into two buckets: months with at least 3 and months with fewer than 3 of such days.  Convincingly, the average 1, 2 & 3-months subsequent returns for months with at least 3 days of top quartile capitulation probability were 0.6%, 1.9% and 2.5%.  Conversely, the subsequent returns for months with fewer than 3 top quartile days were 0.3%, 0.2% and 0.9%.  That said, this analysis supports our thesis that paying attention to days, when perception of market capitulation is high (i.e. frequency of what our statistical work identifies as highly probable “true capitulation days” is high), has historically been rewarding.

Lessons for Near-Term Market Dynamics

What do these results suggest for future market direction?  According to our analysis, May and June stood out as months with unusually high frequency of highly probable “true market capitulation days”.  In these months, we have respectively recorded 13 and 7 days with top quartile capitulation probability relative to both historical and past year medians of 3.  But furthermore, May 20th was also flagged as a “true capitulation day” according to our screening process.  We believe that such strong anticipation of capitulation has ultimately helped fuel the relief rally we saw in July.

At the same time, risk premia across assets has been on decline since May.  VIX has halved to 23 vol points since the year-to-date peak of 46 vol points recorded on May 20th.  As Dominic Wilson has recently noted, the compression of risk premia across asset classes has been justified by the fact that that the heightened concerns about sovereign defaults (in Spain in particular) and broad financial systemic risk have been excessive.  As our view on balance remains that systemic risk and asset market volatility are overpriced, this might suggest that overall risk sentiment has bottomed back in May, which in turn could take downward pressure off equities in due course.

In conclusion, despite some arbitrariness in the definition of capitulation, and the model risk around our results, our empirical analysis corroborates a view that the market may be forming a bottom, and if that is the case, volatility could come down further. However and more broadly, a sustained S&P rally beyond the short-term horizon that this research piece has focused on will likely either require better than expected news on US slowing or a more accommodative US policy response, which is extremely constrained in dealing with a slowdown at this juncture.

And for those curious, here is Goldman's top trade track record: 2 out of 7 is pretty damn good... as long as one does the opposite of what Goldman recommends.

  1. Stay short S&P 500 Dec10/Dec11 Forward Starting Variance Swap, opened at 28.20, with a target of 21, now at 30.38. [loss]
  2. Stay long Russian Equities (RDXUSD), opened at 1645.9 for a target of 2050, now at 1631.95. [loss]
  3. Stay long GBP/NZD, opened at 2.29, with a target of 2.60, now at 2.1606. [loss]
  4. Close short 2-yr GBP swap rates vs. long 2yr AUD swap rates on a 1yr forward basis, opened at -268.5 bp, for a potential loss of 24 bp (inclusive of carry). [loss[
  5. Close short 2-yr TRY rates through cross-currency swaps, opened at 8.77%, with a target of 12.0%, for a potential loss of 168 bp (inclusive of carry). [loss]
  6. Close long 5yr credit protection in Spain vs. short 5yr credit protection in Ireland at 13 bp, opened at 70 bp, with a target of 20 bp, for a potential profit of 2.9% (inclusive of carry). [profit]
  7. Stay long the GS FX Growth Current, opened at 103.5, with a target of 111.8, now at 104.45. [profit]
  8. Stay long PLN/JPY, opened at 32.1, with a target of 37.5, now at 28.4298. [loss]
  9. Stay long Chinese Equities (HSCEI), opened at 12616.01 on 01 Apr 2010, with a target of 15000, now at 12140.5. [loss]