Short-Term Mean-Reversion Becoming Stronger: The Why

This is a (long overdue) follow up to my first post showing that day-to-day stock market mean-reversion is becoming stronger. Understanding this shift is key to trading short-term swing strategies like RSI(2) or adaptive daily follow-through.

In this post, I want to look at why, or more specifically, why mean-reversion is accelerating at warp speed right now - so, put your thinking caps on.

click to enlarge

Recall the graph above from the first post, showing a six-year moving average of next-day returns on the S&P 500 following an up day (blue) or down day (red) from 1950, de-trended to remove the influence of bull/bear markets.

Note how the market today is contrarian, up days tend to be followed by down days (and vice-versa), and that this tendency is accelerating.

The next graph shows the same data as the one above, but for the ETF SPY, rather than the S&P 500 index, from 02/1993. For this post, I need to use the opening values available on the SPY, but not (accurately) available on the index.

click to enlarge

Note how in the ETF, some of the advantage of daily follow-through has been traded out (which is yet another reason I trade leveraged mutual funds, rather than their ETF counterparts), but that day-to-day, the ETF is still very much contrarian and, like the index, this tendency is accelerating.

(Side note: Some might say that we’re talking about very small advantages here (ex. +/- .09% daily on the ETF) and they would be right, but remember that small advantages compounded over time, result in large gains. A 10% annual return is only .04% per day, a 20% return only .07% per day, etc.)

The Why

I would speculate that the biggest reason for the acceleration in daily mean-reversion is the increasing tendency of the market to reverse in the overnight market.

The graph below is calculated the same way as the previous ones, but rather than showing follow-through the next day (close-to-close), we’re only looking at follow-through in the overnight (close-to-open).

click to enlarge

Clearly, daily follow-through in the overnight market made the same shift around the beginning of the decade, and this contrarian tendency continues to strengthen. Just looking purely at the raw average numbers, over half of daily mean-reversion is a product of this overnight mean-reversion.

The Other Why

I would also speculate that part of the reason for the acceleration is that follow-through from the overnight market to the subsequent daytime market (which has historically been consistently contrarian) has now (on average) disappeared. In other words, the daytime market has not been impacted by the overnight market in a consistent way, in more recent history.

Think about that a moment. A couple of years ago, a close down in the markets was (on average) followed by bullish bias in the overnight, but a bearish bias in the subsequent daytime market that moderated some of the overnight change. That daytime moderation hasn’t consistently occurred recently.

To illustrate, the next graph is calculated the same way as the previous ones, but shows follow-through in the daytime-market (open-to-close) from the previous overnight market (close-to-open). I’ve cheated a bit and switched to a five-year average, because it makes the data a little clearer.

Note how average daytime returns following an up versus down overnight have basically moved to zero.

click to enlarge

For reasons that are beyond the scope of this post, I think that this particular shift is temporary, will revert to normal, and will lead to a bit of a weakening in day-to-day mean-reversion. Users of RSI(2), adaptive daily follow-through, etc., beware.

Putting It All Together

So in summary, day-to-day, the markets today (unlike the not so distant past) tend to be contrarian, as up days tend to follow down days (and vice-versa). A large part of that contrarian tendency is the result of mean-reversion in the overnight market, which is becoming stronger.

Prior to the last year or so, some of that mean-reversion was (on average), moderated by the daytime market, but more recently, that moderation has become inconsistent. But for reasons that would make this post much too long, I think that that particular effect is temporary.