Testing the RSI(2) Strategy: Follow-Up

Below are two follow-ups to my recent post applying the RSI(2) indicator to the S&P 500.

First, here is a quick review of my last post: since the late 1990’s, when RSI(2) has fallen below 10/or has risen above 90, it has portended bullish/bearish S&P 500 returns the next day. See graph below from 2000 of next day returns after RSI(2) fell below 10 (green) or rose above 90 (red).

click to enlarge

[logarithmically-scaled]

Follow up #1: Other lookback periods

The first follow up comes from Damian of the quant-oriented Skill Analytics blog.

Damian looks at using shorter or longer lookback periods to calculate the RSI [i.e. RSI(1), RSI(3), RSI(4), etc], for both long and short trades. I recommend reading his post, but I’ll spoil the ending: applied using the very simple approach we’ve taken here, RSI(2) is the most predictive flavor of RSI.

Follow up #2: Other long/short thresholds

In my previous post, I used a threshold of “below 10” as bullish and “above 90” as bearish, but there’s nothing magical about those numbers. Below I’ve tested other bullish/bearish thresholds.

For a long position:

click to enlarge

And for a short position:

click to enlarge

As the above results show (especially for long positions) the deeper RSI(2) has gone into oversold (i.e. low) or overbought (high) territory, the more predictive it has become, but the less often it has triggered a trade.

Based on that observation, a better approach might be to scale positions in or out as the RSI goes further overbought/oversold. In a follow up post, I’ll show the results of such a scaling strategy. I’ve run the numbers already and here’s a teaser, it’s good…dare I say, very good.