- Image by Thomas Hawk via Flickr
On a recent post on Marketsci on the use of RSI2, Bill Luby of VIX and More posted the following comment:
Count me among the fans of RSI (2). I think you might get more interesting results — though with much more time out of the market — using 95/5 and 98/2 as break points.
Also, if everyone is jumping on the RSI (2) bandwagon, perhaps it is time to spend more time evaluating RSI (3) or RSI (4) strategies. Just a thought.
Nice work, as always, but why do I always feel like I am assigning homework when I comment here…?
Well, I agree Bill – you can’t keep assigning homework to poor Michael. I mean, where is the rest of the collective quantitative testing blogosphere when you need it? Michael’s doing all the heavy lifting! So to help Mike out, I’ve done some testing on different RSI days to see what would happen. I can’t promise the snazy charts of Mike’s blog, but I’ll just break it down here quickly.
I created a basic system that buys the SP-500 when the RSI(N numbered day) goes below 10 and sells when it goes above 90. You could do a further optimization of the actual buy and sell levels – I haven’t bothered to do that here. I next optimized the system by having it run through different N numbered days. So, RSI(1), RSI(2), RSI(3), etc. Here’s the results:
So, as you can see, RSI(2) seems to do a lot better across a number of metrics. These include Expectancy, Sharpe and a few others.
Now, what if we altered the RSI days for both entry and exit – meaning, say we enter on RSI(3) < 10 but exit on RSI(4) > 90. How does that look?
So once again, we see that RSI(2) seem to win across a number of factors. Will this hold true in the future? Who knows. Couldn’t you break it down further by period? Absolutely. But maybe I’ll leave something for Michael to do – the guy is so friggin’ lazy. 😉
Good work D!
you might want to look into differential performance of the RSI strategies during bull/bear markets which can be demarcated by using a long moving average such as the 200d MA or the 450d MA (my preference)
there is a big difference between the two, and obviously in bull markets, longer hold times are preferred…….ie 4-5 days, whereas in bear markets, any long position would be closed out in 1-2 days with perhaps a differentRSi sell threshold.
this is because of the speed of rallies in bear markets, and the fact that they reverse quickly. Also as a note, short RSI systems work phenomenally well in bear markets too but are of little use in bull markets.
your thoughts?
I’d agree – it’s funny you mention it because I’ve been running the system using a similar measure – the 200dma – and then looking at ways to “tune” the system to the market. So you can do as you highlight – change the sell criteria during bear periods, or only take one side of the trade. The variations are fairly endless. Thanks for the great comment.
Nice work. Well done.
the other variation i just tested was a volatility filter for the last 8-10 days which was ranked on a percentile basis with a lookback of 100 days……….eliminating the bottom 40% drastically improved returns per trade in both bull and bear markets. Filtering for even more volatile days made the returns obviously higher than that. Of interest was that on days with low volatility in bull markets, short RSI trades got absolutely hammered——–and in general returns were very inconsistent on low volatility days, perhaps indicating that the RSI readings were simply random noise, rather than stretch.
Interesting idea David – I get the idea of the volatility filter – namely eliminating trades where the vol is too low, but I don’t understand what you mean by the filter you actual mention. Can you provide further explanation?
just use an oscillator or percentile ranking of the volatility over the last 5, 10, or 20 days vs the average……..basically you want to eliminate days with below average volatility. compare taking trades using the above strategies when volatility is above average vs below average……..perhaps you may display the table for your other readers——
As a general not the comprehensive system would also look at the 200d MA and now you a very clear picture as a trader of whether to have a long or short bias, and when to bet more than normal—my theory and experience would dictate the following adjustments to the RSI2 strategy:
1) a long bias when market is above the 200d MA–all short trades would have smaller bet sizes and more extreme RSI triggers that would have to be used with intermediate OB/OS indicators
2) a short bias when the market is below the 200dMA–all long trades would have smaller bet sizes and more extreme RSI triggers that would have to be used with intermediate OB/OS indicators
3) bigger bet sizes on more volatile periods
4) cash or trend based systems on less volatile periods (ie 20d or 50d MA) — of course the ADX could help make the determination as to which you would prefer.
as a sidebar the exits preferred by tradingmarkets.com in their research is the 5 day moving average and RSI2=65
Both make sense David – I’ll play around with them and report back.