on the site i saw the crabel stuff on daily data but not ORB ... or did i understand something wrong???
In my opinion, ORB is not a viable daytrading strategy for S&P's, except in a crazy bull market. Of course, in that type of market random long entries will also be profitable. I'd love for someone to prove me wrong, but you'll have to do it with stats, not opinion. I respect Murray as one of the most knowledgeable systems guys around, yet he can't produce a profitable day trade system either. Let the abuse begin.
I agree with you at times. There are times that volatility is too low to daytrade, but you need to look at other markets like Nasdaq and Russell or even Natural Gas. Another issue is how much filtering do you do. If you want to daytrade using the opening range breakout type patterns in the SP500 you need to limit it to 4-5 entries a month , using filters which filter out trades with a lower profits. If you do that then ORB works well. You also need to look at slippage , during the period 1998-2003 you needed to use $300.00 slippage and commission on a big contract to be realistic. about $50.00+ commissions on a mini. The meant you needed to average about $600.00 a trade to be acceptable. The question is if you can use $100.00 slippage and commission now $400.00 would be the same profit. Remember volatility goes two ways. The problem is that except for the Russell , opening range breakout systems are not performing on the stock indexes holding for a day trade or over 2-4 days. The key in trading the stock indexes is to build many different patterns and know when to use which patterns.
i am not sure if volatility is the issue - i guess trend is more important, which is what A-Beltway said (if i interpreted correctly) - i read more than opinion stating that equity indices are among the least trendy markets, and it makes sense: s&p is comprised of sectors some of which are naturally negatively correlated so this bipolarity hinders trendiness.. in the backtesting i have done, i have seen these phenomenal vertical equity curves for day-trading breakouts in s&p from 1982 until around 2002; followed by a stutter.. - having looked at stock market data from 1930s till present i have no reason to believe that it's prudent to expect any time soon a repeat of the kind of trend we had in this last bull market. and i agree with your point about the different markets; in fact there much better trends in individual stock sectors too. Murray - i have a question for you - how do you adjust for volatility in your backtesting? - i typically use 10-20 day historical volatility to normalize my strategy's returns across the backtest period, i am wondering how others do it, thx; all the best.
It depends on if I want to normalize to long term or short term levels. I have use 10,20,40,80 days periods to normalize my results.
This is part of the research process in developing stock index systems and why you can't just test them over a year or so and think they will continue to work in the future. I have seen the same pattern , except for one condition which we inverted be profitable for both the long and short side. That one condition actually caused the pattern to switch modes. That is the nature of the stock indexes. I sorry if I am sounding cryptic, but this is as far as I want to go in this area, because this research is very valuable. I have consulting clients who are paying for this information.
Murray, I'd be interested in knowing your thoughts on how including many filters leads to a more curved fit system. Or is this not an issue since the rules as to how one takes a trade does not change, but what changes is whether the trade is taken or not. I've been trying to understand this issue for a while, i.e whether this is overly curve fitting or not. So any thoughts would be appreciated. Richard
The first and most important issue relating to systems and curve fitting is the premise (the theory) behind the pattern or entry. If that is solid then curve fitting becomes less of an issue. You need to understand the why a given pattern works. Some patterns have easy to explain premises, like intermarket analysis. Other patterns need to analyze based on mass psychology and game theory. There are common structures, for example buying a pullback in a long term up trend. One thing which helps make sure filters are not a curve fit is to see if the filters by themselves are predictive? They don't have to be tradable, just give you a reasonable positive expectation. Also testing individual filters means you have a lot of cases. This curve fitting issue is very complex and I will address it in my on line-system development course starting here on ET in a week or so.