TTO’s Modified RSI(2) and the “Best 2-Day Indicator” Redux
This post was originally going to be a test of Trading the Odd’s modified RSI(2).
But it got me thinking (again) about how to best objectively test this myriad of two-day indicators (RSI(2), DV(2), etc.) and really say which is the “best”. I think my last attempt was analytically-sound but the conclusion a bit murky. I want one set of data I can point at and say, such-and-such is the best at trading so-and-so.
And so, I give you the best 2-day indicator redux…
Note: these results are for the S&P 500 index (or more appropriately, things that track the index closely like leveraged mutual funds). I’ll look at SPY in a bit.
The five indicators under the microscope are: RSI(2) calculated based on the S&P 500 index, RSI(2) based on SPY, TTO’s modified RSI(2), DV(2), and a Stochastic %K.
All of these indicators share two things in common: first, they all range from 0 to 100, with 0 = completely oversold, 50 = neutral, and 100 = overbought. And second, as they move further away from 50 the signal becomes stronger (ex. 10 is more oversold than 30).
So here’s the test…
Assume five fictional portfolios. For each go X% long/short at each day’s close depending on how “deep” the indicator is into oversold/overbought territory.
An indicator value of 50 = no position, 40 = 20% long, 30 = 40% long, etc. This way we’re not just rewarding the indicator for getting the next day’s direction (up/down) correct, but also for focusing exposure on those specific days that are the most oversold/overbought.
Geek note: results are frictionless (i.e. no transaction costs or slippage).
Results
I’ve used four metrics to judge the “best”. The first three are self-explanatory. The last one (“consistency”) deserves a post all to itself, but in a nutshell, I’m capturing how consistently the indicator performed over the entire sample (high = good, low = bad). My thinking has evolved on this and these results are different than any other “consistency” metric I’ve talked about before (more on this in a future post).
And the winner is…Index RSI(2).
DV(2) was a close second, but results were not as consistent across the entire sample.
Trading SPY
Like any short-term strategy, these five are very sensitive to the specific asset traded, so below I’ve rerun the test on the ETF SPY.
And the winner is…DV(2).
TTO’s modified RSI(2) was a close second, but came in just behind DV(2) on all metrics.
One last bit of data to chew on…
There is of course no reason why multiple indicators couldn’t be combined to confirm each day’s reading, so below I’ve included a correlation matrix of daily indicator values.
I especially like how our two winners, index RSI(2) and bounded DV(2), exhibit (relatively) low correlation to one another (making them potentially even more useful in unison).
Last Thoughts
As the stats show, all five indicators have been effective trading both the S&P 500 index and ETF since the turn of the century, and as the correlation matrix shows, all have given similar reads on the market on any given day.
But if I had to choose just one to best capture the short-term state of things it would be index RSI(2) for the S&P 500 and DV(2) for the SPY.
Happy Trading,
ms
. . . . .
To stay up to date with what’s happening at the MarketSci Blog, we recommend subscribing to our RSS Feed or Email Feed.
Filed under: Trading Strategies | 8 Comments
>these five are very sensitive to the specific asset traded, so below I’ve rerun the test on the ETF SPY.
Curiosity: I wonder how these would perform for example for the RUT.
Would they give better or worse performance, what do you reckon?
RE to George: good question…possible subject for another blog post.
Really two questions I think: (a) running all of these same indicators using S&P 500, but trading RUT and (b) running all of these indicators using RUT and also trading RUT. michael
I would come to a different conclusion, Michael. The fact that DV2 was at the top in risk adjusted return for two different series suggests more robustness. I guess I will have to await your consistency measure to see why it trumps this observation. Love your blog.
Jerry
RE to Jerry: you make a valid argument, but I think that the Index and the SPY have shown such significant differences in performance with short-term strategies (daily follow-through being a very good simple example) b/c of minute differences in daily closes, that I don’t think it makes sense to aggregate the results from both. I really see them as two unique entities (but only for very short-term strategies), each with a “best” answer.
Having said that, there hasn’t been enough difference between RSI(2) and DV(2) performance on the index to say with absolute certainty which is the best, and I think either conclusion would be perfectly reasonable.
More to follow on the consistency metric.
michael
How about a normalized version of RSI(2)? DV(2) has the advantage of being normalized to the empirical distribution of the index rather than just make an assumption about whats oversold/bought. Franks RSI is pretty much the same as the DV(2) but it´s not normalized. Maybe it´s the normalization that adds to DV(2)s success over the other indicators.
Nice post!
RE to Emil: great comment – on the todo list to test. michael
A linear combination of RS(2) and DV(2) (e.g., summing them and investing the result) won’t offer any additional advantage over dividing one’s account into two parts and using one on each part. But what about taking a position only if they agree on direction? In that case trade the average of the two.
One would think there should be other ways of combining them that aren’t linear, but off hand I can’t think of any.