Technical analysts spend many an hour agonising over the best entry techniques and the best systems for trend following. But, frankly, is all the effort worth it? My testing proves not; random entries are as good as any.
Ed Seykota’s ‘Whipsaw Song’ contains his rules of trading in its lyrics:
- Ride your winners
- Cut your losses
- Manage your risk
- Use stops
- Stick to the system
- File the news
What if a trend-following system follows all six of these rules but uses random entries to enter a trade? Can such a method possibly be profitable?
If random entries can produce a profit then it ought to give trend following practitioners a great deal of confidence in the basic tenets of good trading set out above. And it is surely a great deal more difficult to over-fit such a system to the data.
The system: I set up a simple trend-following system to test random entries as follows.
Entry: If there is no position in a particular instrument, take one: either long or short on a random basis. Random numbers are produced from 1 to 20: if 1 is produced by the random number generator, then a long position is taken, if 2, then a short position. For any other number no action is taken.
While there is no position in an instrument, this process is repeated daily until a position is taken. Thus there are periods in which there is no position in a given instrument. As can be appreciated, by increasing the random number range (to between 1 and 50 by example) you can force longer periods of abstinence and less trades overall in the portfolio.
Exit: A trailing 5 ATR stop based on a 20-day simple average ATR. When a position is stopped out it will eventually be re-entered either long or short as per the entry rules above. ATR is average true range – a measure of the recent volatility of an instrument.
Risk management: This consisted of limiting the initial position size on entry. On entry 0.375% by value of the portfolio will be risked (volatility adjusted, fixed fractional position sizing based on the distance to the stop). No other attempt will be made to limit risk.
Portfolio: The portfolio consisted of a balanced assortment of over 25 futures, 3 bonds, 3 currencies, 3 energy, 4 grain, 3 interest rates, 3 metals, 2 softs and 3 stock indices.
Test runs: I ran 1000 tests. I could have run 5000 or 500,000 but I suspect the overall statistics would differ little.
- No interest earned on unused cash balances
- Slippage: 7%
- Commission per contract: US $7
- Start Date: 1st January 1990
- End Date: 1st February 2013
- Starting Capital: US $40,000,000
Average statistics on the 1,000 tests run came out as follows:
- Percentage of tests profitable: 100%
- CAGR (compound annual growth rate – i.e. this is the return): 2.42%
- MAR (average CAGR divided by average draw down – a pain-to-gain ratio): 0.19
- Max Peak to Valley Draw Down (biggest loss in portfolio value at any period): 14.11%
- Number of trades: 1,995
- R Squared (smoothness of returns): 83
- Standard deviation (annualised monthly): 5
- Winning trade duration (days): 145
- Losing trade duration (days): 52
The conclusion is that almost any form of trend following has worked very well for the past few decades. I seem to have demonstrated that provided trends actually exist, almost any entry method will work so long as combined with an exit which lets profits run and cuts losses short.
Further tests showed how the results can be improved by such measures as adding a long-term filter, so that trades are only taken if in the direction of the long-term trend. When such a filter is applied, the aggregate results very much resemble the historic returns and risk-reward ratios of the trend following CTA community as a whole, emphasising that exact entry methods are not important. The industry as a whole is no better at picking entry points than my random system.
The game has undoubtedly become more difficult over the years for a number of reasons, including the increasing number of players in the markets. The years of 2001 and 2012 proved exceptionally tough and few trend following systems (random entry or otherwise) were able to profit in such choppy and trendless markets.
What does the future hold? Who knows, but it is as well to be prepared. If strong and long-lasting trends re-emerge, such systems will once again profit. If not, they won’t. It’s as simple as that really.
My object in testing random entries coupled with a trend-following exit was a very deliberate one. To test whether trend following works on such market data as I have at my command.
My object was not to test the efficacy of random entries coupled with random exits, which even intuitively I realised would be a zero sum game. Nor was it to test whether random entries (with or without trailing stops) would be profitable or otherwise on a sideways market – again, even intuitively I realised they would not be.
The experiment was to test trend following on a broad base of markets, where almost by definition there would be strongly trending periods and periods of sideways on/off movement, deadly to long-term trend following.
Anthony Garner is author of A Practical Guide to ETF Trading Systems.