What would be a good filter for a Mean-Reverting system?

Discussion in 'Strategy Building' started by Oadmani, Sep 10, 2021.

  1. Oadmani


    I am in the process of making a mean-reversion trading strategy. The current winning percentage I am getting is 67. However, the strategy tends to incur many, many consecutive losses in periods of sustained downtrend. For instance, during the 2008 financial crisis, it would probably have 20-30 losses in a row. If I were to implement the strategy as is, I would get almost wiped out in a sustained downtrend.

    Anyone mind sharing a way to filter out trades during periods in which markets collapse?
    murray t turtle likes this.
  2. 2rosy


    only trade the system when volatility < N
    Gambit and Oadmani like this.
  3. Oadmani


    Good point. What would be the best measure to assess volatility in the context of a sustained downtrend?
  4. SteveM


    "If the VIX closed > X yesterday, do not take any trades today" could potentially help.
  5. You first need to study mean reversion as a factor.

    1. Start with a hypothesis -- what happens if you buy a stock that dips below 5% from it's 5 day moving average? (complete random example)
    2. Collect the data on subsequent returns through 1, ..., x days etc.
    3. Decompose your returns by overlaying other bits of information: volume, historical volatility, information about the stock (is it a small cap, large cap, etc.), was there an event (e.g. earnings), etc.
    4. Use your decomposed returns to find what subset of mean reverting opportunities leads to higher (positive) subsequent returns
    5. Measure stats on a historical backtest (standard deviation, sharpe, mean returns, etc.) to figure out how much $ you can make and if you can lever up on the strategy, and what kinds of tailrisks you need to hedge
    6. Run it with a live account (with a small portion of capital) for 6-12 months
    7. If all works well, add more capital
    shuraver, toon and Oadmani like this.
  6. panzerman


    67 percent is a very good winning percent, but does it translate into positive expectancy? The more conditions you add to your basic system, the fewer trades it will generate with no guarantee of being positive expectancy. You will have to do some careful backtesting to determine that.

    In general, mean reverting strategies fail during sustained trends. IMO there is nothing that will work to avoid that truism, just limiting the loss is the the best you do. Be sure to backtest. No positive expectancy, no trade. Sorry to be a downer, but I suspect you won't find such a system.
  7. lindq


    Start with VIX<30 as a filter. You'll miss many profitable trades, but it may help to keep you from ending up on the street.

    You can also try combining VIX levels with different signal levels for mean reversion.

    But a word of caution. If you are testing equities looking for long signals, and using lows on daily bars as entry points, daily lows from data providers can often be suspect. You'll need to do a lot of forward testing to confirm your expectations.
    yc47ib, jnbadger, d08 and 1 other person like this.
  8. zghorner


    envelopes could be useful for this strategy.
    murray t turtle and Oadmani like this.
  9. 2rosy


    you could research market making logic. some inputs are
    • volatility
    • holding time
    • current position
    • desired position
    • liquidity
    • risk tolerance
    murray t turtle likes this.
  10. %%
    QQQ filter ................................. ''Almost wiped out''; trading to BIG.
    Maybe can get by with it if your younger+ can afford to blow up an account; '' if it works/ add more capital'', especially when SPY gets aboVe 202dma, like most of 2009. To blow up faster add more capital in a 2008 bear market ..................................................................... To blow up faster risk 3.33or 6.66 %. Some make a mint almost risking 7-8% but thats trading with main trend+ seldom a winning 70% hit rate:caution::caution::caution::caution::caution::caution:,:caution:
    #10     Sep 10, 2021