Skip to content

SMR- A Sentiment and Mean-Reversion Trading System for Bear Markets

August 18, 2009

Going long during bear markets can be a very scary proposition. Extreme RSI 2 or DV2 readings give us little consolation when peering over at the abyss. I wanted to design a system that incorporated a variety of factors to maximize the probability of success, and minimize the probability of an extreme loss. I also wanted a system that would be in the market for as little time as possible. I call the system “SMR” because it combines  sentiment indicators along with  mean-reversion indicators. I define bear markets in this case as the SPY being below its 252-day average. Here are the system inputs:

Mean-Reversion Indicators

1) unbounded DV2

2) daily follow through

3) DVS (the stretch indicator)

Sentiment Indicators

1) the VIX- volatility index

2) Gold and Silver — via the Central Fund of Canada (CEF) a favorite of true gold bugs

SMR System Rules:

It is a bear market: SPY is < 252-day average of closing prices

1) The DV2 unbounded is less than zero

2) The market is  lower than yesterday’s close

3) The DVS is less than the 50th percentile

4) The SPY outperformed CEF since yesterday’s close

5) The VIX is higher than yesterday’s close

If all of these rules are all satisfied buy the SPY on close, sell when the criteria are no longer met (usually trades last 1 day). Here are the results for the last 2 years:

SMR System 23/08/2007- Present

 
 
  Avg Gain % Correct Max Loss Max Gain # Trades  
Buy Signals

1.7%

80.2%

-4.40%

14.50%

25

 
             

SMR System 23/08/2007- Present Performance Summary

 
 
  CAGR StDev Sharpe R-Square

DVR (Sharpe x R-square)

 
Buy Signals

22.0%

13.4%

1.59

88%

1.40

 

System performance is very good with an annual compound return of 22%, and the equity curve is very smooth with an R-square of 88%.  The average hold is less than 1.1 days, and the maximum loss is only -4.4%.  There were only two trades that triggered two days in a row, and they were both winners.

9 Comments leave one →
  1. bill kreyenhagen permalink
    August 18, 2009 1:28 pm

    Could you show some of the dates that this triggered so I can make sure I have my charts setup correctly?

    • david varadi permalink*
      August 18, 2009 1:48 pm

      hi bill, here are a few recent dates: 08/07/2009, 06/07/2009, 16/03/2009, 09/03/2009

  2. toptick permalink
    August 18, 2009 4:20 pm

    Very interesting. Thanks!

    Readers should note that the ‘pre-sample’ results (94-07) are not too bad, either.

    • david varadi permalink*
      August 18, 2009 4:27 pm

      thank you toptick, i just checked……roughly 70% correct and nearly 1% per trade over the last 3000 bars

  3. August 18, 2009 5:47 pm

    David,

    Very interesting indeed. Have you done some testing in bull markets (by your 252MA definition)?

  4. scrilla_gorilla permalink
    August 20, 2009 3:36 am

    May want to consider incorporating IG vs. HY bond divergences as another measure of risk appetite… for example relative daily change in LQD vs. JNK. I believe when the bond prices diverge from the equity movement (i.e. JNK outperforms LQD when SPY closes down), the bond prices have predictive value for the next day’s equity movement.

  5. August 20, 2009 7:56 am

    Why don’t you use z value as a performance indicator. It’s like a sharpe ratio that separates luck from consistency. Z value above 3 with out stop loss included would be quite a strategy. Do you use Z value?

    • david varadi permalink*
      August 20, 2009 10:02 am

      hi, im aware of using the z-value, i try to report measures that i believe most readers will understand or grasp intuitively. With regards to your second point, i acually use a percentrank function to combine these factors which is superior to z-values as it does not assume a normal distribution. You will see this method being used in previous posts on various indicators. So i do agree that it is a useful method, its just that there are better methods available. The “z” would be my second choice.

  6. August 21, 2009 10:30 pm

    That is very interesting because I heard about the Z value for strategy test, the normal distribution was the 1st thing that came to me. I’ll look for that percentrank function that you mentioned. Sounds interesting to me. thnx

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: