Skip to content

Iron Chef Battle Oscillator –RSI2 and DV (SPY)

March 29, 2010

In case some of you missed the announcement, we just realeased DV Indicators for Amibroker. The first package released contains mean-reversion/oscillators with several new indicators. The standard DVO, DVB (DV Bounded),  DVSC (DV-Supercharged), DVI (DV Intermediate Oscillator) and DVDS (Super-Smoothed Double Stochastic) and a few others are also in the same package.

The recent market period has been largely a momentum market, so I think its a good to time to highlight how the DV bunch have fared in the onslaught versus the RSI2 using the SPY. Note that this is not a response to the good analysis done by Michael at MarketSci http://marketsci.wordpress.com/2010/03/22/roundup-the-best-2-day-indicator/ which is an attempt to quantify which indicator is the most reliable from a consistency standpoint at extreme levels on the ^GSPC or S&P500 index over a long period of time. That particular post was directed more towards trading funds as the S&P500 index is not tradeable.  The only two instruments that are tradeable for most traders are the SPY and e-mini S&P500 contract.  Rather this post is more about what is working right now in the midst of what is arguably a much less favorable market for short-term trading and mean-reversion. The tables below show a brief summary of the frictionless returns to trading the RSI2 using 50/50 and 30/70 levels for entry and exit as well as the different DV variants using the median for the entry exit threshold. All indicators were traded both long and short. The numbers speak for themselves—RSI2 has been taken to the cleaners in relative and absolute terms whether the benchmark is buy and hold or the different DV variants. All DV variants beat buy and hold by a large margin. The king of the DV oscillators the DVO shows the best performance. Note that the original DV2 was privately conceived in a virtually identical form in mid 2008 as a tool for my own pairs trading, so this is effectively all out of sample performance! (MarketSci was kind enough to bring DV2 to the attention of the public in mid 2009). 

Advertisements
14 Comments leave one →
  1. Frank permalink
    March 29, 2010 5:07 am

    Hi David,

    in 2010 DVSC2 and DVO have exactly the same performance metrics. This must be a misstake…?

    Frank

    • Frank permalink
      March 29, 2010 5:15 am

      …. sorry forgot to add: Thanks for this post!

      I can definitely confirm your results as i’m trading using DVO as well.

      All the best….

      Frank@DE

    • david varadi permalink*
      March 29, 2010 12:12 pm

      hi Frank, it is not a mistake the DVO and DVSC had the exact same trades this year.
      thanks again frank–cheers
      dv

  2. John French permalink
    March 29, 2010 10:50 am

    David, is there somewhere that gives a brief overview of the indicators that you now offer? Every time I look there are more! Re the Excel version will it be an add-in a la Ta-lib? Will there be documentation?

    I have coded DV2, DVO, DVS, AggM, AggZ, DVTRSI from your site and the DS Oscillator from Corey’s but have not looked at them that closely thus far (next on my list after taxes). Are any the same as the retail versions or is the latter deemed to be universally superior where there is an overlap?

    Basically I want to prototype some ideas using what I have thus far and then purchase the Excel version if I get decent results from my testing.

    Thanks

    John

    • david varadi permalink*
      March 29, 2010 12:16 pm

      Hi John, I will in fact be coming out with documentation—thanks for pointing that out. the DVO that was mentioned on the site was a framework–not the actual version that is for sale. the DVS is actually the DV Stochastic and not the DVDS which is what you have probably coded (the DV Super-smoothed double stochastic). the DVS is a short-term stochastic and bears no relation. the AggM and AggZ are not in this package–some of the new ones such as the DVTO and DVTS have never been mentioned before.

      I will be sure to send you any new documentation in the future John.

      best
      david

  3. March 29, 2010 1:47 pm

    David,

    Could you tell us the number of trades each indicator made during this test period?

    Al

    • david varadi permalink*
      March 31, 2010 7:58 pm

      hi al, i don’t have the test handy but I can find out. please email me at dmvaradi@gmail.com

      best
      david

  4. Jack permalink
    March 31, 2010 7:19 pm

    For those of us without TradeStation of Amibroker, are the Oscillators and Mean-Reversion Indicators available for sale in either their raw equation form, or coded in Excel?

    I’m very impressed with the performance of the simple, unbound DV(2), which Michael Stokes (MarketSci Blog) made available as an Excel spreadsheet, and would like to see the other variations on that theme.

    • david varadi permalink*
      March 31, 2010 7:56 pm

      hi jack –thank you very much. an excel platform will be available next week that will have adaptive indicators and other variants not available through other releases. this will also contain all dv indicators and several new ones as well.

      best
      david

  5. April 12, 2010 9:55 pm

    That’s very interesting, but nobody has ever suggested using RSI2 like that. I don’t think it would have worked well at any point in the past. Most people sell when RSI2 is greater than 90, 95, or 98, and buy at corresponding levels (10, 5, or 2). There’s really no point in using RSI2 with trades at 50, because you will get almost identical trades just by selling every up day and buying every down day. This comparison is just as badly flawed as the one that MarketSci did, it’s just flawed in a way which favors different indicators.

    If you really want a good comparison of the indicators, you should use a rolling rank on all of them. The best value for overbought and oversold is always changing, and the ranking will correct for that. In my testing of various oscillators, I’ve found that the ranking lookback period makes a bigger difference than which indicator is used. Almost all of the fast oscillators have correlations well over .8, so they just aren’t different enough to be statistically significant.

  6. david varadi permalink*
    April 12, 2010 10:46 pm

    well, clearly unlike the marketsci post this was partly a marketing post so i will credit you that. however, in reality what you are saying is flawed because its like saying that all ETFs are the same because they are correlated. all academics understand the momentum effect is the most robust anomaly in the stock market, and that holding the strongest ETFs risk-adjusted delivers alpha. there is such thing as momentum in indicator performance that is tradeable, and the DVO outperforms anything i have seen or tested that is public—it has done so for 10 years by a very large margin. im not sure how else to argue with that other than to say that at extremes there are differences in how these oscillators behave as you suggest. however i don’t see why you wouldn’t just use the most “reliable” or “best performing” indicator for that task. Personally a 37% cagr over 12 years is a hypothesis test that says: what are you going to use on a day to day basis that outperforms this simple measure?
    best
    dv

    • April 13, 2010 7:12 pm

      Actually, there are two main things that I was saying:

      1. Nobody trades RSI2 at 50 or 30/70, so the results from trading at those levels don’t really matter. RSI2 is normally only going to be profitable if the trades are taken at extreme readings, like 2/98, 5/95, or 10/90.

      2. Using a rolling rank instead of fixed overbought/oversold levels makes a bigger difference than the details of how an oscillator is constructed. If you take the formulas for your indicators and remove the rolling rank part and trade it based on fixed overbought/oversold levels, I would expect the performance to be worse than trading RSI2 with a rolling rank to determine the overbought/oversold levels. If you really want to show that one indicator is better than another, you should use a rolling rank on all the indicators.

Trackbacks

  1. A Simple Adaptive Change to Daily Follow-Through to Capture Cycle Changes « CSS Analytics
  2. DVI system (1/2) « Quanting Dutchman

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: