Improving your trading strategy using probability

Discussion in 'Strategy Development' started by Eddy, Nov 4, 2002.

  1. Eddy


    I just have been reading on a (french) trading website the following suggested way of improving the % winner of any standard trading strategy :

    Just say you have backtested a trading strategy and it shows that you have 45% of winners.
    Based on that strategy, your create a new trading system :
    you follow your initial system in realtime but you wait that it produces two consecutive losses, before entering on the following entry trading signals that it is triggered... According to some calculations (I don t know how the following percentages are calculated), you should increase this way your avg odds of getting a winner from 45% to 75 % (and it would even increase to 83% if you wait 3 consecutive losers, 90% for 4, 95 % for 5, etc)
    Another suggested option to benefit from the "higher odds" of a winner after a series of losers would be to take all trade generated by the initial system, but to increase your bet size after consecutive losers (and decrease it after a series of consec winner...)

    Ok, that s what I have been reading : it looks like this is an easy way for increasing your system profitability, but i have no clue (even some doubts) if such a filtering would really work...

    The percentages provided sound logical (ie you have more chance to hit a winner after 5 consec losers than after a winner), but I remember back from my university time that one should be very careful with probability calculations... Of course, taking a trade after two losers will reduce the number of trades of the filtered system and I don t have no idea of what will be the new Ratio avg Win/avg Los....

    I intend to do some backtesting runs to see what kind of result this produces on my favorite trading systems... I already backtested in the past several stop loss / profit target strategies (fixed stop / volatility related, etc) and many trend indicator filters, but never tried to filter yet the trades I would take depending on the PL of the previous trades...

    In the mean time, if anyone could come with some feedback about such a method (ie personal experience on that, some book reference, or even some older ET thread that I may be bypassed), this will be very much appreciated

  2. dottom


    What you are doing here is creating a system to trade your system. Specifically in your case you are creating a position sizing system to manage your system. (Some "systems of systems" completely discard a signal after X winners or losers.)

    It is a valid concept, but you need to make sure that you have a statistically valid amount of trades before you draw any conclusions.

    For example, many trend-following strategies have a higher chance of the next X trades being losers after a large win. This has to do with the nature of trends. Is the frequency high enough to enable increased profitability by using a system to trade your system? Only with thorough backtesting can you make a decision.

    Similarly, many counter-trend systems show an increase chance of a winning trade after several losing trades. This is a general statement so I should add a large "your mileage may vary".

    But the big question remains whether the correlation between prior and subsequent signals is large enough to take advantage.
  3. I cannot attach much credence to this advice. Either you trade your system, with its inherent underlying probability distribution, or you don't. This sort of manipulation, ceteris paribus, will not skew the odds in your favor. On the assumption that you have no crystal ball, the money saved by missing out on those 2 losing trades will, in the long run, be exactly offset by those frustrating occasions that you watch from the sidelines all the winning trades going by as you wait for your 2 consecutive losing trades.
  4. Aaron


    If you flip a coin and get tails four times in a row are you more likely to get a head the next flip? Of course not, each flip of the coin is independent of all others and has a 50% chance of getting heads.

    Your trade by trade results are also probably independent and there is no reason to increase your size after a randomly occuring losing streak. In fact, you'd be better off decreasing your size to minimize your risk of ruin.

    Position sizing and probability are lengthy topics. You should research "martingale bet sizing" and "auto-correlation".
  5. acrary



    That's a pretty clear example of the Gambler's Fallacy. The problem with the thought process is the trades have to be shown to have some dependence for this to work. While regression to the mean is a viable concept, the odds don't improve on any single trade (unless they are dependent).

    Here's a link with more information on this:
  6. this is a ridiculous approach to statistics. If you were able to foresee in advance a trend such as two winning trades / two losing trades / two winning trades, you wouldn't be playing around with statistics would you, you'd program your system to do just that. Unfortunately, the market is RANDOM (why do people always try to curve fit a random occurence). Granted there are familiar set ups, but as you say they don't always pan out the right way.
    You will lose good trades and enter bad trades just as often, even by waiting for those so-called bad trades to happen. I can almost assure you, that your win/loss ratio will not CHANGE a bit if the population remains the same.
    Good Luck.
  7. dottom


    Disagree 100%.

    Research on trend-following methods clearly show a higher % chance of losing trade following larger winning trade; and higher % chance of winning trade after series of losing trades.

    Also, read some of P2's threads. His entire trading method is based on increasing position size anticipating reversion to mean.

    Look at spread trading techniques which rely on reversion to the mean.

    However, I will say that generally speaking position sizing techniques should only be considered after you are capable of trading one method consistently using standard money-management techniques.

    But to say that position sizing techniques is the same as betting progressions in random coin flips is naive. It is an advanced topic and should not to be pursued lightly or could have detrimental effects on your trading (both method and psychology).
  8. This is vague at best. The market is random; under what time frame? All time frames? Over the past century, the overall trend for the DOW has been up. This obviously isn't concise proof that all markets are non-random, but it most certainly doesn't help your case either.

    The past two years, the Nobel prize for Economics has been awards to teams that established a basis as to why economic systems exhibit non-distributed behaviors ( 2001,2002 )
  9. This is an interesting concept. Very similar to the martingale betting system (if you're playing blackjack for example). When you are playing basic strategy black jack (no counting), the odds are very close to 50-50 for each hand (with a slight edge for the house). So if you lose 5 hands in a row, those are almost the same odds of flipping 5 heads in a row for a coin. The odds of losing 10 hands in a row are extremely small ( I believe around 1 in 1,000).

    Although with the coin flipping story, it could be argued that if you flip 10 heads in a row, over the very long run, the odds are you will flip less heads than tails on subsequent flips. I realize it is suppose to be "random", but then again it is suppose to be 50-50 odds over the very long run (reverting to the mean).

    Stock prices in one sense are random, yet in another sense are not random. For example, I guarantee that the DOW JONES won't go up for 365 consecutive days. It never has and it never will. There will be up days and down days. There will be swings in sentiment. Those you can be certain about. The randomness comes in the actual price movement. Will the DOW be at 8876.64 on January 1st or 9560.20 or any other number? I have no idea.
  10. dottom


    This is Gambler's Fallacy. It can clearly be proven that the results of each coin flip is indepdenent. Any observation showing "reversion to the mean" over the long-term is just standard probability distribution.

    Do not confuse mean reversion as it applies to trading to gamblers fallacy.

    Disagree. The market is a complex system, but definitely not random.
    #10     Nov 5, 2002