### RSI(2) vs. DV(2)

In this post I want to flesh out the differences between tried and true RSI(2) and a new indicator I introduced yesterday DV(2) from friend of the blog, David Varadi.

The graph above shows the results of two trading strategies applied to the S&P 500 ETF SPY, versus buy & hold in grey, from 2000 to present.

The first (blue) is an extreme RSI(2) strategy that goes long at today’s close if RSI(2) closes below 10, and goes short if closes above 90. The second (red) is an extreme DV(2) strategy that goes long if DV(2) closes below 10, and short above 90.

*Geek note: this is a proof of concept so these results are frictionless (i.e. do not account for transaction costs, slippage, or return on cash). Also, I am not purporting that these threshold (10/90) are the best (in fact, they’re very binary which I am an opponent of), but I’m trying to keep things simple and compare apples to apples.*

And for the number lovers… note that I’ve also included trading a 50/50% blend of the two as well:

At first glance of the performance graph, it would appear that these strategies are more similar than they actually are. Only about a fourth of the time that at least one triggered did the other trigger. The other three fourths of the time, only one or the other signaled a trade.

This opens up an opportunity I think to analyze divergences between the two, or at least gain some level of diversification of short-term strategies (note reduced average drawdown and improved risk-adjusted returns in table for 50/50% strategy).

**Other Thoughts…**

First, RSI(2) tends to only work well at extreme readings such as the 10/90 tested here. DV(2) does the same, but also works pretty well at less extreme readings (above or below the midpoint) which we saw in my previous post.

Second, even though both indicators are using similar lookback periods (2 days), they are using very different data sources. RSI(2) relies purely on close-to-close price changes, while DV(2) is looking at the close relative to the day’s trading range. Again, an opportunity to analyze divergences or at least trade in parallel (which, as I talked about in Same Strategy, Different Data, can lead to improved risk-adjusted performance).

Third, I’m not cheerleading for DV(2). It’s an interesting addition to the toolbox, but as I preach incessantly (and recently gave a bit of a negative book review because of), at the end of the day, trading success or failure is driven by much bigger concepts (ex. confidence-based strategies, trading in multiple timeframes, etc.)

*[Edit: click for a **summary of all posts** in this series on the DV(2) indicator]*

Happy Trading,

ms

*P.S. A big thank you to David for sharing his work with us and allowing us to pick it apart in a public forum.*

*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 | 11 Comments

How about using a dynamic oversold/overbought measure on these strategies? For example by using bollinger bands?

Sure…I think there’s all sorts of neat things that could be done with this indicator (but that’s outside the scope of what I usually post about…I just give a vanilla test of the indicator and then let readers torture it to their hearts content =)

michael

Great post!! How many buy and sell signals did both strategies generate? Also, how many days on average were you in the market after the signal was generated?

Thanks.

RE to Raj: thanks for the kind words. 314 with an avg. hold of 1.9 days for RSI(2) and 309/1.7 for DV(2). michael

On a lark I tried Ver 1 (unbounded) of DV(2) for SPY, IWM, and DIA for their full histories. My results were positive, but not exceptional. Two notes/questions:

1. For DV(2) I am averaging today’s DV and yesterday’s DV(2). This may be different than your calculation (which is slightly ambiguous). Do you have this 2-day running average, or is your calculation DV(2) = 0.5*DV(today) + 0.5*DV(yesterday). I suspect that may lead to some of my underperformance.

2. In my version (running avg. DV(2)) shorts almost always did poorly – either flat or slightly losing, except in the last year and a bit of extreme market collapse. Have you looked at a long-only version of this?

I’ve been impressed by what I’ve seen on you site so far.

RE to Zack: I’ve had quite a few people a bit confused over my less-than-robust calculation description. See this xls file: http://www.marketsci.com/supporting.docs/dv2_calculation.xls

Strategy #1 and #2 in the file are the unbounded DV(2) above/below 0, and the bounded DV(2) below 10/above 90 strategies.

michael

Michael, yet another great post. Thanks for the ideas to shake and bake.

hi Michael,

in my past efforts at testing these, i had noticed that some of these applied to broad ETFs like SPY seemed to work far better in high volatility environments. Your graphs seems to bear that out. they work better in the bear markets (characterized by higher volatility) vs. the bull period between 2003-2007.

comments ?

as always, congrats on the high quality work !

regards

Two immediate thoughts…(1) even if win % was constant, it will always appear that a strategy is more effective in a high vol. environment (more money to be made on daily swings), and (2) I didn’t test the following statement on this particular strategy, but generally speaking, I’ve found short-term mean-reversion indicators to work at least reasonably well long in all environments (bull, bear, and sideways) but short only in sideways and bear markets – I think this has something to do with the market’s tendency to slowly grind upwards, but fall very quickly. michael

thanks for your comments and insight..

i’ve had to use more medium term momentum strategies (RSI being too short term) for the bull markets, intermixed with short term based on RSI and such for the bear market areas to get a better bang for the buck in the past.