Trader Effects

An observer effect is a concept from physics in which the act of measuring a phenomenon sometimes affects that phenomenon. The observer disturbs the experiment by the act of observing. A similar thing happens in trading: The act of trading itself can change the underlying market conditions on which the success of a trade is predicated. I call this a trader effect. Anything that repeats with enough consistency is likely to be noticed by several market participants.

Similarly, a strategy that has worked especially well in the recent past is likely to be noticed by many traders. However, if too many traders start to try to take advantage of a particular strategy, that strategy will cease working as well as it did previously.

Let us examine this concept using a specific example. Assume that there were no buyers willing to buy at $408.00 or higher but there were sellers willing to sell 1,000 contracts at anything above $409.00, and these sell orders would act as a ceiling keeping the price from going over $409.00. Before the addition of your buy orders, the market would not have advanced to the price of $410.50, and so the breakout would not have happened. Therefore, in a simulation for a breakout-based system that looked at this trade there would not have been a breakout and thus no trade.

Now suppose that in the same circumstances you enter the market and buy up those 1,000 contracts at an average price of $409.00; there are no more sellers at that price, and so you have to buy another 100 from the sellers at $411.00. This trade causes the large buyer to begin buying, at which time you sell him the 1,000 contracts at $411.00. Although he thinks he got a good price, you made an excellent trade. All that remains is to get rid of the extra 100 contracts. Since there are no buyers at the recent prices, you have to sell lower, and so you sell your 100 contracts back where the market had been trading: at $407.00. You lose $4  100 ounces on 100 contracts, or $40,000, but you made $2  100 ounces on 1,000 contracts for a new profit of $160,000 not counting commissions. Not bad for a few seconds’ work.

Since you would be buying and then quickly selling right afterward, you might change the meaning of a breakout itself. Before the addition of the trader effects, a breakout might have signified that resistance had been broken, and so there was a greater likelihood of favorable price movement when a breakout occurred. However, with the addition of the new buys, which are designed only to move the market enough to cause a breakout to occur, the meaning of the breakout has been altered.

I’m sure that one of the reasons for that system’s sudden unprecedented drawdown was exactly this sort of anticipatory buying, which effectively ruined its edge for a time. They will exploit any repeated patterns that they notice. This is one of the reasons why it is better to develop your own system; you can build a system with which it is much less likely that you will have your edge ruined by trader effects because other traders will not know exactly when you will be buying or selling.

When we traded for Rich, there were often times when we would all be entering trades at essentially the same time. Market traders knew that when they started to get large orders from us, the orders probably would continue for quite some time. For that reason, at times the floor traders and brokers would start to move the market ahead of our orders. Since we used limit orders, this was a bit more risky—that was one of the reasons we used limit orders—because we would not get filled in those circumstances and so we could pull our orders. Sometimes when I wanted to buy and knew that the market was particularly prone to having the locals move it in anticipation of our orders, I would send fake orders in the opposite direction. Then, if the market moved, I would cancel the original order and place a limit order much closer to the market or even on the other side of the ask.

For example, if I wanted to buy 100 contracts, I might first place a fake sell order. If that fake order was a sell of 100 contracts at $415 and the market was trading at $410 bid and $412 ask, the presence of the order might move the market to $405 bid and $408 ask. I then might cancel the fake limit order and enter one to buy at a $410 limit that probably would get filled at $408 or $410, which was the original ask before my first order. In some respects it was a bit like bluffing in poker.

You cannot bluff all the time or you’ll get called and end up losing the bluffs and your bets. However, an occasional bluff can help your play considerably because it forces the other players to call you sometimes when you actually have a good hand, resulting in a larger pot when you have the winning hand. You also may win pots with the bluff, and that also increases your winnings.

Random Effects

I created a complete system that combines random entries based on a coin flip with a time-based exit some number of days later within the range of 20 days to 120 days. I then ran 100 tests with the same data that was used in Chapter 10 to compare trend-following strategies. The best test in the simulation returned 16.9 percent and turned $1 million into about $5.5 million in the 10.5 years of the test. The worst test in the simulation lost 20 percent per year. This shows that there is a good deal of variance that is due entirely to random events.  What happens if we add a little edge?

What happens if we add a little edge? What if we make our system similar to a trend-following system by including the trend filter we use with the Donchian Trend system so that trades are
entered randomly, but only if they are in the same direction as the major trend?

If you add a trend filter with a positive edge to the completely random system, the average performance for 100 tests moves up considerably. In my simulation, the average return rose to 32.46 percent and the average drawdown dropped to 43.74 percent.

Consider the best track record from the 100-test simulation that was cited above. If one traded less aggressively, for example, at 25 percent of the level we did as Turtles, one of the tests would have achieved returns of 25.7 percent with a drawdown of 17.7 percent with a 10-year track record. We all know that a trader who entered randomly would not be more likely in the future to perform at the same level since there is no edge in trading randomly. Unfortunately, for anyone looking only at a track record, out of a large group of traders some will have been lucky enough to seem to know what they are doing when they actually do not.

What happens to variance in our test if we use a shorter time frame, perhaps only the 3.5 years from January 2003 through June 2006? For this period, the average performance for the random entry system was a 35 percent return with a MAR ratio of 1.06. The real systems did considerably better.

  • The Triple Moving Average system returned 48.5 percent with a MAR ratio of 1.50.
  • The Bollinger Breakout system returned 52.2 percent with a MAR ratio of 1.54.
  • The Dual Moving Average system returned 49.7 percent with a MAR ratio of 1.25.

As for the random tests, how many lucky traders emerged from the 100 tests in the simulation? How many beat our best system’s performance purely on the basis of luck? Seventeen out of 100 had a MAR ratio better than 1.54; of those 17 tests, 7 had a return higher than 52.2 percent. The very best random trader returned 71.4 percent with a drawdown of 34.5 percent for a MAR ratio of 2.07. All this is something to think about the next time you are looking at a three-year track record with excellent performance.

The Optimization Paradox

The optimization paradox states that parameter optimization results in a system that is more likely to perform well in the future but less likely to perform as well as the simulation indicates.

Consider the Bollinger Breakout system, which has two parameters. Figure 11-1 shows a graph of the values for the MAR ratio as the entry threshold parameter, which defines the width of the volatility channel in standard deviation, varies from 1 standard deviation to 4 standard deviations.

Note how the results for a channel with a width of 2.4 standard deviations are the peak for this simulation. Any value for the entry threshold that is less than 2.4 or greater than 2.4 results in a test that shows a lower MAR ratio.

Now, returning to our premise that optimization is beneficial, suppose we had not considered optimizing the channel width and instead had decided arbitrarily to use a channel width of 3.0 since we recalled from high-school statistics that 99-plus percent of values for normal  distributions fall within 3 standard deviations of the average. If the future is fairly close to the past, we would have been leaving a lot of money on the table and would have subjected our trading to much greater drawdowns than a 2.4-standard deviation entry threshold would have provided. To give you an idea how great that difference could have been, consider that the test at 2.4 makes 8 times as much money over the 10.5-year test with the same drawdown as the test at 3.0 does, with returns of 54.5 percent versus 28.2 percent for the test with an entry threshold of 3.0.

We know that if the channel width is too narrow, you get too many trades, and that hurts performance; if it is too wide, you have given up too much of the trend while waiting to enter, and that also hurts performance.