Testing RSI Momentum Before and After Publication
Summary
The document considers whether publishing a technical indicator can reduce its subsequent trading performance. It frames a strategy’s historical results as a mix of genuine alpha and overfitting, with only genuine alpha expected to persist out of sample; wider adoption may also cause alpha to decay. This suggests that an indicator-based strategy could have performed better before it became widely known.
The author tests a 21-day RSI trend-following strategy on 32 commodity futures from 1960 to 2016, using prior-day information and roughly equal risk allocation. Gross results are stronger before the indicator’s 1978 publication than afterward, and the strategy is described as flat from 2009. These results are suggestive rather than proof of a publication effect: the analysis excludes transaction costs and notes that historical computing access and trading feasibility may differ. It also uses RSI as a momentum signal, unlike its original proposed use as a reversal indicator.
Key ideas
- A strategy’s in-sample performance may combine genuine alpha and overfitting.
- The RSI trend-following backtest performed better before its publication than afterward.
- The test uses commodity futures, a 21-day RSI lookback, and roughly equal risk allocation.
- The results are gross of transaction costs and may reflect historical execution constraints.
- The test uses RSI for momentum, although its original proposal used it as a reversal signal.
Tags
Full text
# How well would Technical Indicators have done before creation?
# How well would Technical Indicators have done before creation?
Backtesting technical indicators (that determine market psychology) - how well would they have done?
For example RSI was published in 1978. Before its creation - did it performed better or worse?
## Answer by Chris Taylor (score 4)
https://quant.stackexchange.com/a/35159
A simple model for any trading strategy that is fitted using historical data is that the in-sample performance is some combination of
- Genuine alpha
- Overfitting "alpha"
and that out of sample, you just get the genuine alpha. Additionally, alpha may decay over time, but it is unlikely to improve. In the absence of other effects, you would therefore expect any trading strategy to have much better performance before it was published and became widely known.
To test this with RSI, I used a data set of 32 commodity futures from 1960 to 2016, and created a simple trend-following strategy that uses the RSI indicator with a 21-day lookback period. Trades on day t are formed using information up to and including the close of day t-1, and are assumed to be executed at the closing price on day t. Positions are sized to give roughly equal risk in each contract over the entire sample. All results are gross of transaction costs.
The results pre- and post-publication are
```
1960-1978 1979-2016
Annualised Return 8.79% 3.54%
Annualised Volatility 5.97% 5.49%
Sharpe Ratio 1.47 0.64
```
It is clear that the strategy has not performed as well since it was published, and in fact it has been completely flat since 2009 - indicating a possible post-publication effect.
Of course, other explanations should be explored (e.g. running a systematic strategy with daily rebalance may not have been feasible before the widespread availability of personal computers, or transaction costs may have been prohibitively high in the past).
One final note - I have used the RSI as a momentum indicator here. The original article, published in Commodities magazine by J. Welles Wilder in July 1978, envisioned the RSI being used as a reversal indicator, i.e. you would sell when the RSI was high and buy when it was low. Given the strong results from using it as a momentum indicator prior to publication, I am at a loss to explain this!Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.