Ways of testing Mean Reversion

Discussion in 'Strategy Development' started by etfarb, Mar 25, 2013.

  1. etfarb


    How could you test mean reversion of a price series without getting all wild with something like cointegration?

    I was thinking about making a moving average, then determining the spread of the ma and the price series, then find the STdevs and figure out parameters for when it gets to extended or reverts back to the mean.

    Can it be done this way or do you need to use cointegration?
  2. dom993


    Markets have essentially 2 modes, trend or chop.

    - In chop (trading range), your MA is a good proxy for that "mean".

    - In trend, what's happening is that the market consensus on price moves away from its last agreed-upon area (trading range), and it actually moves away faster than price do (for most of the trend, anyway). So in a trend, that "mean" is ahead of the current price levels, and your MA is terribly lagging until market establish a new consensus on price, which creates the next trading range, were that MA will settle as a valid proxy for that mean.

    I am not a specialist of pair trading, but my understanding is that the whole point is to find pairs which ratio have the least chance of getting into trend mode.
  3. vikana

    vikana Moderator

    I think there are two types of analysis that are valuable. You already mention some calculation of deviation from a trend/mean using stdev() or some other measurement.

    An often overlooked one is serial correlation, i.e. how do one bar relate to the prior bar(s). That's often interesting and will help you determine trending characteristics.
  4. Daring


    Markets typically revist most moving averages, but they have a higher chance of doing so when the moving average, within itself was trendless.
  5. so how do you find the best way to see the switch from trend/mean or vice versa..
  6. Now, THAT, is the $64,000,000,000,000 question.
  7. right.. +1 on that..
  8. If want to do some type of reversion strategy, you want an ideal series to have stationary properties (hence co-integration/pairs, adf tests, etc). By taking the reversion from the moving average of a single series, the results can be very undesirable (think about what happens when a series is continuously increasing, and you bet short on a price that stretches far above its moving average -- even if it reverts back to the ma, your short has no reason to be profitable, as the moving average is also likely increasing).
  9. If you follow the Intuition Amplifier thread patiently, you may find it starts to unravel the solution and a lot more.
  10. what about TPO moving mode?
    #10     Mar 29, 2013