'Long S&P 500 Overnight' Is Not an Advisable Strategy in Bear Markets

Last week Goldman Sachs analyst Peter Berezin published a report revealing that a strategy that had bought the SPY at the close of trading each day and had sold it at the following day’s open would have returned over 300% since 1993.

In an article entitled “Bull by Night, Bear by Day is Best Strategy, Goldman Sachs Says,” Bloomberg quotes Berezin as follows:

A large number of market participants are averse to holding overnight positions, which causes them to sell at the close (thereby depressing intraday returns) and buy at the open (thereby inflating overnight returns)…. Such aversion to overnight risk is likely to be higher during bear markets…. Even when one is pondering less trading intensive strategies, the analysis above suggests that there is a cost to be paid for avoiding overnight risk.

While we don’t have access to the original report and therefore may be taking the original argument out of context, the implication appears to be that traders should act against their natural risk-aversion instincts and be long the index overnight (potentially even more so during bear markets, when the “aversion to overnight risk is likely to be higher”).

But is being long the index overnight really a prudent strategy? Even in bear markets? To answer these questions, let’s first reproduce Goldman’s results.

Figure 1 shows the compounded return (excluding transaction costs) since 1993 from buying the SPY at the close of each day and selling it at the following day’s open.

Adjusting for dividends, the strategy actually appears to have returned approximately 400% since 1993. Despite this total return, however, it’s clear from the figure that there have been two extended periods during which the strategy has had quite painful drawdowns. Upon inspection, it’s hardly surprising to discover that these two periods happened to coincide with the only two bear markets within the testing window, namely, the bear market from 2000-2003 and the present bear market that started in late 2007. The peak-to-trough drawdowns in each case were -40% and -27%, respectively.

As a result, we have to question the advisability of (unconditionally) holding the index long overnight while in the midst of a virulent bear market. To be fair, the strategy has performed superbly during bull markets, which suggests that if we could only distinguish between bull and bear markets dynamically through time, we could adjust our positions accordingly (for example, by holding short positions overnight during bear markets).

A Better Strategy?

To this end, we propose a simple dynamic adjustment to Goldman’s strategy—an adjustment that would have helped to ameliorate the strategy’s severe negative performance during bear markets. The adjustment is simply to condition the polarity of our position (long or short) on whether the SPY is above or below some measure of its long-term moving average.

Figure 2 shows the (frictionless) results of several strategies that go long or short SPY overnight depending on whether SPY is, respectively, above or below its 200-, 300-, or 400-day moving average (along with the benchmark strategy of holding long overnight).

As one can see, the simple adjustment of conditioning one’s position on whether the index is above or below its long-term moving average, to a large degree, smoothes out the two major periods of underperformance for the original strategy. Table 1 lists the performance and risk statistics of the benchmark and adaptive strategies. These statistics further support the case that conditioning one’s position on the market’s dominant trend adds value. In particular, note the difference between the benchmark strategy and the adaptive strategies when it comes drawdowns and recovery time (i.e., the time needed to recover what was lost in the drawdown).

Table 1	Benchmark	200D MA	300D MA	400D MA
Total Return	444%	617%	885%	886%
Winning Trades	2072	2041	2070	2072
Losing Trades	1588	1619	1590	1588
Winning Pct	56%	55%	56%	56%
Average Return	4 bps	5 bps	6 bps	6 bps
Standard Deviation	66 bps	66 bps	66 bps	66 bps
Sharpe Ratio	1.00	1.25	1.50	1.50
Average Win	40 bps	42 bps	43 bps	43 bps
Average Loss	43 bps	42 bps	41 bps	41 bps
Win/Loss Ratio	0.94	1.01	1.03	1.03
Max Drawdown	-33%	-14%	-13%	-17%
Recovery Time	794 days	96 days	93 days	97 days

(For a more in-depth discussion on the utility of using long-term moving averages in asset allocation, see Faber.)

An Enhanced Strategy?

We now seek to further enhance the strategy by considering two additional factors, both of which take into account the mean-reverting nature of the S&P 500. These two factors are:

a) “The Law of Reversion” factor

b) 1-day stochastics

The Law of Reversion states the following: For any time series, X, that is not overly serially correlated and has median value, M, when X(t) is greater than M(t), there is a 75% chance that X(t+1) will be less than X(t) (and similarly when X(t) is less than M(t), there is a 75% chance that X(t+1) will be greater than X(t)). (See “Statistical Arbitrage” by Andrew Pole for more details.)

How does this apply to the SPY? Well, without completely revealing our proprietary indicators, suffice it to say that it is possible to create a time series involving intraday SPY price differentials that adheres to the Law of Reversion—which is to say we can predict with roughly 75% accuracy whether the next price differential will be greater than or less than the most recent price differential (note this is not the same as predicting whether the next price will be greater or less than the current price—rather it’s more a prediction of volatility).

The second factor that we consider is the 1-day stochastics of the SPY. This factor is defined as:

1-Day Stochastics = (High – Close)/(High – Low)

Why include 1-day stochastics? Well, given the mean-reversion in the S&P 500, when the 1-day stochastics has reached an extreme at either end of the spectrum, historically the odds have been fairly strong that the index will move in the opposite direction on the following day. (For ongoing commentary on the evolving reversionary nature of the S&P 500, please see Michael Stokes’ MarketSci Blog.)

Putting It All Together

Combining the long-term moving average conditioner with these two additional factors, we can construct the following trading algorithm:

If:

• The q₀-day moving average is greater than q₁, and
• The Law of Reversion factor is greater than q₂, then
• Go long at today’s close and exit
  • At tomorrow’s open, if today’s 1-day stochastics is less than q₃
  • Otherwise exit at tomorrow’s close

Similarly (on the short side),

• If the q₀-day moving average is less than –q₁, and
• The Law of Reversion Factor is less than –q₂, then
• Go short at today’s close and exit
  • At tomorrow’s open, if today’s 1-day stochastics is greater than 1 – q₃,
  • Otherwise exit at tomorrow’s close

A distilled version of the algorithm can be thought of as follows:

• The long-term moving average governs whether the strategy will be long or short
• The Law of Reversion factor determines whether the strategy will take any position (provided that position is in accordance with the long-term moving average)
• If the strategy is in the market, the 1-day stochastics governs whether the strategy will exit at tomorrow’s open or at tomorrow’s close

For simplicity and a desire not to overfit, we take the parameters q₀, q₁, q₂, and q₃ to be 400, 0, 0, and 0.9, respectively (similar results hold for reasonable variations of those parameters). Figure 3 shows the (frictionless) results of applying the aforementioned strategy to the SPY, while Table 2 lists its performance characteristics. Both the equity curve and performance characteristics of the enhanced strategy suggest that the combination of trend-filtering and the exploitation of mean-reversion holds promise as a means of timing the S&P 500.

Table 2	Benchmark (1994-2009)	Enhanced (1994-2009)	Benchmark (2004-2009)	Enhanced (2004-2009)
Total Return	444%	960%	25%	163%
Winning Trades	2072	1087	712	349
Losing Trades	1588	700	596	223
Winning Pct	56%	61%	54%	61%
Average Return	4 bps	14 bps	2 bps	17 bps
Standard Deviation	66 bps	84 bps	74 bps	85 bps
Sharpe Ratio	1.00	2.60	0.43	3.25
Average Win	40 bps	56 bps	40 bps	56 bps
Average Loss	43 bps	53 bps	43 bps	43 bps
Win/Loss Ratio	0.94	1.07	0.92	1.30
Max Drawdown	-33%	-9%	-23%	-5%
Recovery Time	794 days	38 days	Not yet recovered	19 days

From Table 2 we see that the enhanced strategy is in the market roughly 50% of the time, has experienced relatively low drawdowns (and recovered quickly), and its performance has improved in recent years—likely due to the fact the S&P 500 has steadily become ever more strongly mean-reverting. The (frictionless) average return per trade of the enhanced strategy suggests it may be viable in practice, if one can consistently execute with low transactions costs.

Conclusion

The takeaway of the foregoing analysis is that blindly holding the SPY long overnight can be exceedingly dangerous in bear markets. Given that we are in the midst of a relentless bear market, now is not the time to be holding long positions overnight. But as Goldman suggests, market participants should not uniformly shun overnight risk. By conditioning their positions based on the dominant trend of the market, traders can likely improve upon a “holding-long-overnight” strategy. Moreover, further enhancements to an overnight strategy may be gained by taking into consideration the strongly mean-reverting nature of the S&P 500 (along with its dominant trend) and designing one’s trading strategy accordingly.

Disclosure: The author is long SDS, DUG, and DTO.

Comments

[David Sawyer:]

Great work here - really like the detail... any concerns as the market could rally back here to 800 on the Spx?

Mar 16 10:04 AM

[morph366:]

Very good article -
Thanks for sharing these ideas and the methodology

Mar 16 10:13 AM

[J. D. Swampfox:]

It seems then, that if it is undesirable to be LONG overnight, then it might actually be desirable to be SHORT overnight in these wild bear market times ...

Mar 16 10:14 AM

[squark62:]

i wouldn't be hold SDS w/o a fraction of SSO. trend is our friend but our friend isn't a magic mirror on the wall (or maybe the reverse depending on how you view the analogy.) even if the remainder of your portfolio is bullish the s&p again i'd still hold a little SSO for a balanced hedge.

Mar 16 10:45 AM

[kotika98:]

there is one little problem with Goldman analysis, and a big one with Raymonds.

I have a problem with Goldman, in that before a couple of years ago, SPY was listed on the AMEX, which means you did not get decent executions, specialists there were notorious crooks. Costs of this strategy would far outweight the benefits in those days. These days SPY is trading electronically so it would not be so hard to execute the strategy, but you get my point. I would check if this conclusions hold if you did it with futures, around 9:30am, and i am not too sure it would still be profitable. We are talking about an average alpha of less than 10 basis points per day here.

The second one is with Raymonds analysis - anyone who is serious about quantitative strategies would know that the way to cure the bear market performance is simply to be long during the night, and short during the day. Thats a very efficient and clear way to avoid the bear market, or in the jargon of the profession to capture "alpha" instead of "beta".

Mar 17 09:55 AM

[Raymond Mic...:]

Unfortunately, Kotika98, long overnight/short during the day doesn't work that well during bear markets. It's ok as a strategy but fairly choppy.

R Micaletti

Mar 17 10:22 AM

[omooc:]

Excellent presentation -- valuable until you introduced proprietary devices, which are not evaluable by outsiders.

Mar 17 10:33 AM