Note: daily signal updates for the adaptive version of this strategy are available in the State of the Market report.
Probably the simplest indicator I’ve ever talked about on this blog is daily follow-through. By “follow-through” I mean if the market is up today how likely is to be up tomorrow, and if it’s down today, how likely is it be down tomorrow?
In my previous study, I showed that this concept has had a consistent impact on the stock market over the last 60 years, but that the direction of the impact has evolved. To illustrate, the following graph shows going long/short the S&P 500 tomorrow (at today’s close) when the market closed up/down today from 1955 (frictionless).
For 40 years, follow-through was consistently positive – down days tended to be followed by down days (and vice-versa). But around 2000, the market made an abrupt change (dotted line) – down days now tend to be followed by up days (and vice-versa).
Adaptive Strategies
The markets are constantly in flux and the themes that work today won’t work tomorrow. And because of that, I’ve made the case that our ultimate goal should be strategies that learn from the markets…that are adaptive.
Below is this same follow-through strategy (blue) overlaid with the adaptive approach (red) plus the abnormal market filter (green) that I use in the State of the Market report.
[logarithmically-scaled]
Note how the adaptive strategy actually trailed the original strategy for quite a few years as it began to detect the change in the market and take less aggressive positions. But as the original strategy began to invert, the adaptive strategy evolved with the change. The graph below of annual returns for both strategies makes this a bit clearer.
Also note that the addition of the Abnormal Market Filter reduced total returns, but increased risk-adjusted returns (see table below) because it reduced portfolio exposure when the market became stretched and (by my theory) unpredictable.
Calculation Notes
For the adaptive strategy, I’ve assumed that the percentage of capital invested each day equaled the 5-year confidence figure that I would have provided in the SOTM report. For the abnormal market filter, I’ve assumed that the trader then reduced that percentage of capital by the SOTM report’s “percentage abnormal” figure.
Note that this is a proof of concept so this study is frictionless (no transaction costs or slippage). Unless you trade leveraged mutual funds from Rydex, ProFunds, or Direxion like I do, trading frictions would impact these results.
Closing Thoughts…
The main points I hope I’ve conveyed in this post are: (1) very simple indicators like this one can be very powerful (remember, we don’t have to be perfect – we just have to compound small quantifiable edges over time), (2) the fundamental characteristics of the market are constantly in flux, (3) static strategies will eventually fail, and (4) adaptive strategies, while not a magic bullet, are a step in the right direction towards permanently wrangling these unruly markets.
Happy Trading,
ms
P.S. some previous posts on designing adaptive strategies: How to Build an Adaptive Strategy, a Simple Example, and Coping with Abnormal Markets.
To stay up to date with what’s happening at the MarketSci Blog, we recommend subscribing to our RSS Feed or Email Feed.
Filed under: Evolving Markets, Follow-Through, Stock Market Mechanics, Trading Strategies | 1 Comment
Mike – Running adaptive strategies on multiple instruments appears to work very well as well.
I am working on some things now where there are adaptive systems running on individual equities & a signal is generated for the upcoming period. These signals are then evaluated & ranked by a risk-adjusted expected profitability. Based on this ranking, trades are placed with the quantity to trade being determined by a money management algorithm.
The idea behind this is that even though there are adaptive systems, each system will experience times when their profitability is reduced & therefore there are better instruments to trade for the upcoming period.
It’s amazing how much one can learn from a system when they throw a data mining tool or do some simple risk/reward analysis & analyze the factors which contribute to winning trades vs. losing trades.
The next step for me is to develop this across multiple asset classes that are uncorrelated & keep it contained so it does not turn into a 3 ring circus. :)
Regards,
Eric