Market Timing - Things to Watch Out For
Market Timing is something I have spent a fair amount of my life attempting to do - that is, to trade in and out of the stock market and yet acheive higher returns than the index (preferably with lower risk). I have found it very nearly impossible, and for the most part have given up my efforts to find such a system, but occasionally I get an idea and will research it, enjoying the challenge but knowing full well my efforts will likely be in vain.I have also monitored the attempts of others in this area, and since the late 90's I have regularly bookmarked the websites of various market timers. I should mention that very, very few stand the test of time. I would say less than 1%, a sobering thought.
What I wanted to share here was some of the warning signs of websites that are more likely to fall in the 99%+ of market timers who end up disappearing.
Outright failure
Kind of obvious, but if a system is losing money for a year in an easy buy-and-hold bull market, that's a bad sign. It is somewhat expected that a market timer may underperform a buy-and-hold strategy in an outright bull market, but actually losing money in such a situation is not good.Lack of Performance Data
Again, should be obvious, but there are some subtle forms of this which I'll get into below. As far as the overall concept, market timers research historical data to spot repeatable tendencies and in this way develop and backtest their systems. What worked in the past is expected to work in the future. But this is a very big expectation. The markets change over time, often due to outside influences. One example is that before trading in stock market futures began in the early eighties, the market tended to close randomly between the high and low of the day. Since that time, though, the market is much more likely to close near the extremes of the day. This affects trading systems that are based on the short term behavior of the market close, and thereby limits the amount of data that can be used in the backtest. When doing this kind of work you want to use as much data as you can (i.e. a large sample size). Within this limited dataset, you would like to see many examples of all market conditions (such as bull and bear markets), but this is generally not possible. Many timers who looked successful in the nineties failed as the market turned in 2000 because their backtested data was comprised of nothing but bull markets.Some sites actually have NO performance data whatsoever. In my mind these should be avoided like the plague.
Lack of Information on Risk
Sometimes you may see a system that seems to have a good return, but they give no information on how much risk is involved in acheiving that return. Surprisingly often you will see systems that have drawdowns of 40% or more. Few people can trade through a drawdown that deep. Most will throw in the towel. A 20% drawdown is fairly reasonable, and certainly less than the stock market itself will regularly experience, but this will cause most people a fair amount of stress as well. You simply must know how much risk the system is likely to have, even if it's only by looking at an equity graph.Unclear where real time trading begins
This is one of the biggest problems with market timing sites. This is a critical piece of information, because it is very easy to optimize a system to perform admirably over historical data, but it is absolutely worthless unless it works in the real world.Good timers will leave a substantial amount of recent data aside (out of sample data) and will test their system on this set. Ideally this out of sample set should include all the types of markets that will be encountered (bull and bear). Few people use substantial out of sample data, and those who do generally end up making it useless by continually checking their out of sample results as they test the model. This is called "data snooping" and usually ends up with the data overfit to the entire historical dataset. Realtime data starts where the out of sample data ends. A very good timer may leave yet another sample to test on, a sample that will be used once. At this point, it either works, or you give up. But few give up.
The bottom line is that you can only judge a system by how well it has done in real time trading, so this is a must-have piece of information, and one that is often hidden by market timing websites.
Model continually changed as it fails
This is a common ploy - the model fails in real time, and the developer changes it so that it would have performed better during the period of failure. This piece of information is typically hidden from potential purchasers, and the re-optimized historical data is presented as actual performance by not specifying where real time trading began. Sometimes the "wayback machine" on archive.org can be helpful in this regard.Not so good when evaluated by a 3rd party
A few timers use 3rd parties like TimerTrac and others to measure their performance in real time. This is a great idea, but it is extremely surprising to me just how often the TimerTrac results are far worse than the results posted on the websites. Just a couple of clicks and you can see how they actually have done, yet I have to assume most people don't bother, as a lot of these sites would go out of business if they did.Numerical Gymnastics
Some sites will use some metrics like the additive total number of points gained. It's not the worst way to do it, as some people will be trading futures based on this information and will not be compounding quite the same way as someone trading SPY, but it often makes the results look better than they would be otherwise. I often will open a site in an Excel spreadsheet and compute the percentage changes and other metrics, and I have seen this kind of analysis make a great point-based system look no better than buy-and-hold.Averaging vs. Compound Return
More numerical hijinks. Some sites will take many years of compounded historical returns and then divide by the number of years to give an average return. This gives a much higher number than could be acheived in reality. A ten year return of 300%, or 3, would reflect an average return of 30% a year, while the actual returns would be computed as 3^(1/10)-1 or an 11.6% compound annual return. Not so impressive anymore.Multiple stocks
What I'm referring to here is something I have only noticed fairly recently. These are the sites that trade multiple stocks a day, which is fine, but they count these performance-wise, as separate, sequential positions, which is about as wrong as you can get. Let's say that today you divided your portfolio into 3, and traded 3 stocks today, all bought at the open and sold at the close, and each one made 5% (nice!). Well, what you would have made that day is 5%. But these sites would calculate that you made 15.76%, which is what you would have made if you could have put your entire portfolio into one stock, made 5%, then gone back in time with the extra 5% and invested it in the next stock and so on. Impossible and overstated results.Leverage
Leverage is often used to pump up performance data. This is legitimate, but it goes back to the problem of risk, and lack of information on risk. You use a 2X leverage fund, and yes you double your return, but yes you also double your drawdown, and my belief is that not many can tolerate big drawdowns, and virtually no systems are good enough to merit the use of that leverage.Also, many do not understand the mechanics of a 2X fund or ETF. These are geared to provide 2X the return for each day, but this is not mathematically the same as 2X the return over a longer period.
Lack of Equity Graph
This really is the case of a picture tells a thousand words. Also, the graph needs to be logarithmic, as an arithmetic graph of compounding equity will skew the results. On a log chart a straight line represents a constant compounded return, and dips in the graph (drawdowns) can be compared accurately.