Evolving Markets and Dynamic Systems: Daily Follow-Through
I’ve been devoting a lot of time over the last year to the idea of systems capable of adapting to evolving markets rather than being coded with static rules. For example, our original MarketSci strategies are very static and require manual tinkering as markets change. By contrast, our new YK strategies have no hard-and-fast rules – rather, they are designed to “learn” how to trade from the market.
In this post, I want to look at a very clear example of the market evolving over the last 60 years or so: “daily follow-through”. What I mean by follow-through is if the market is up today, how likely is it to be up tomorrow, and if it is down today, how likely is it to be down tomorrow?
That’s a simple enough question to answer, but here I want to show how the answer has been changing over the last 6 decades. The graph below shows a 5-year moving average of the next-day returns on the S&P 500 following an up day (red) and a down day (blue) from 1950 to 06/2008.
Geek notes: (1) returns have been normalized by subtracting the average return of all days in each observation period to remove the influence of bull vs bear markets, (2) averages are geometric.
For most of the last 60’ish years, the market has exhibited at least some degree of follow-through; up days tended to follow up days and vice-versa. This peaked in the first half of the 1970’s when there was a +0.43% difference between next-day returns following up vs down days
However, since the mid-1970’s, this difference has been slowly eroding, and since the new millennium, it has actually flipped to contrarian. Up days now tend to be followed by down days, and conversely, down days tend to be followed by up days. As of the end of June, the five year average difference in performance was about 0.16%.
Now, absent other factors, the difference as of this moment isn’t large enough to be tradable; however, it is large enough that traders might need to consider it as a part of their overall strategy.
But more importantly, it demonstrates that the fundamental characteristics of the markets are in constant flux. We must adapt to the mood of the markets and be ever open to changing our assumptions about “the way things work”.
Happy trading,
ms
Filed under: Evolving Markets, Follow-Through, Stock Market Mechanics | 9 Comments
While I would guess that quant black box systems are enforcinging the very short-term mean-reversion for us, I would guess that trends are no less strong in longer timeframes, as one can hardly (at least in this point in time) stop the business cycle and mass-crowd psychology. For this reason, at some point of timeframes of diminishing size, follow-through must be present. I would venture to guess that the size of the timeframe is just ever so slighly longer than that of the next day. Therefore, I’m curious what the follow-through is when comparing the aggregate return for three-consecutive days is, for example, when compared to another three consecutive days, only starting one day later. In other words, you will be comparing the return of the first non-overlapping day to the return of the other non-overlapping day (three days later). I bet there is still follow-through there.
RE to LyricalOne: I’ve covered this quite a bit on the blog – at longer timeframes the market does trend, but in short/intermediate-term time frames it is (at this moment in the market’s history) contrarian. These concepts underly the performance of our YK and Scotty strategies. michael
I would suspect this is b/c of more and more trading at shorter intervals (more swing trading in the 70′s, more day trading in the 90′s, etc.). Interesting how things seem to both change AND stay the same all at the same time.
MT :)
Yes, I understand. I’m just curious at what point follow-through goes into effect, and mean-reversion leaves (2 days after the day used as the benchmark, 3 days, etc?)
RE to LyricalOne: of course, it depends how you define follow-through (i.e. are we looking a down days, multiple down days, moving averages, etc.) but here’s one study that looks at your question (and how it has changed historically):
http://marketsci.wordpress.com/2009/01/13/the-moving-average-spectrum-part-i/
http://marketsci.wordpress.com/2009/01/18/the-moving-average-spectrum-%e2%80%93-part-ii/
I also stumble upon the “filp” (from continuation to mean-reverting) in US indices data. Note that smaller indices stills exhibit daily continuation. The turn is around year 2000s, which I suspect has something to do with decimalization of the quote.
In my mind, it works something like this. In a quoting system that has big tick size such as in the US pre-decimalization, it is more costly for trader to move a price. Therefore any buy/sell decisions must be backed by a firm believe. In other words, you can’t buy at the ask then turn around to sell at the bid immediately. Think of a poker table where there is a high minimum raise. Then players wouldn’t use “probe” bets as often.
Thanks for a good blog.
P