Here is a backtest of talon's pullback idea from 2003 up to now. The universe of securities is the most liquid NYSE or NASDAQ stocks refreshed every 6 months (no survivor bias). I took all the signals that came out of talon's rules and sized the position of each bet to have a $100 dollar daily volatility (according to my forecast of the volatility at the time). I allowed for 5bps slippage off the close (5bps in and 5 bps out) which a professional trader should find very generous for this universe of stocks. To address the fact that in the backtest I used the close to make a decision to trade on the close, which is a bit forward looking (of course in practice, a mid quote snap around 3:39 should be good enough and you are able to send MOC orders), I generated a backtest where all the trading decisions are delayed one day (therefore perfectly actionable). The equity curve strictly following the rules should lie somewhere between the 2 backtest.
Forgot to say, the universe above is the 500 most liquid names on NYSE or NASDAQ refreshed every 6 months. Talon, what don't you like about Larry Conners research?
Most traders overemphasize it. That is one reason there is so few longterm, lucrative traders. The problem with most traders, is they do not have a real edge and/or their money management is not at a professional level. Even a depressed neurosurgeon can do successful operations. He has the knowledge and the skill. Discipline is wrapped up in top money management, position management, portfolio management in concert with the edge.
Yes, I agree with this... except a depressed trader usually cannot trade... there is something about performing at the very top level that requires a person to be well as a whole being.
Thank you for doing this backtest and presenting this equity curve. Having not tested these exact parameters on your data set, I cannot exactly confirm your results, but I can say it looks "about right". One feature of equity curves from all of these types of systems (buying/shorting pullbacks on a portfolio) is these weird sharp drops. You also should look at how many positions you have on at one time... I would imagine in a universe of 500 stocks you could easily have 300 positions on at some times. Why? Because stocks are highly correlated. This makes you consider how much money you really need to trade this and what risk is involved... and how to manage that risk. Still thinking on that other question from yesterday. It's been a long day... will write more tomorrow most likely. Thanks again. Larry Conners - Everything he does is one system and I don't think his testing procedure or statistical approach is sound. Just my opinion.
Talon- Thanks for your efforts here - appreciated. I was actually going to ask how you manage the correlation risk on a basket of stocks within the same strategy. I'll take a crack at it and the corr risk between multiple systems, but another question came to mind first. When you're ready to trade a newly developed system on a portfolio of stocks, do you consider the performance of each individual stock in your testing? For example, you test on 500 stocks, and 50 of them consistently do not make money across periods - do you take them out of the portfolio for live trading? This relates to the corr question - whether or not it makes sense to consider the historical performance of individual stocks when looking at position sizing / trade priority.
Curious, why only one position at a time? Due to market correlation? What about if one position was short and one was long at the same time?
Good question - this refers to each individual stock. You could buy and have a buy signal 2 days later when you're still in the first trade. You would not take the second signal, but if you're trading 100 stocks you theoretically could have 100 positions on at the same time.
One additional question if you don't mind. For a system that has multiple day holding periods, is this as simple as taking the system's average log return per day in the market and getting its percentile rank against all the daily log returns over the entire period? Where a percent rank of 50% would be no better than random and >75% would be good?