I'm skeptical... what would you test next?

Discussion in 'Automated Trading' started by bigmoetx, Feb 11, 2022.

  1. bigmoetx

    bigmoetx

    Hope everyone is having a great 2022 so far.

    I've typically traded over longer timeframe (swing trading to investing timeframes) but have recently caught the algorithmic trading bug (again).

    It has always been a side goal of being able to develop something more automated - on shorter timeframes.

    In the past, I've had a hard time developing a tradable system. Common issues include:
    (a) They don't have positive expectancy :)
    (b) Even when they have positive expectancy they don't have a tradable equity curve, e.g.,
    i - Deep drawdowns or long multi-year valleys
    ii - too few trades to be statistically significant
    iii - Miniscule avg trades
    And so on.

    About a year back, I had an hypothesis/idea that I thought was worth exploring. After some time evaluating the idea - exploring stats of underlying price movement - I developed a system and have been back-testing.

    Here are results from an out of sample backtest of about a year across basket of stocks:

    Out of Sample Results (~12 month window)

    Annualized return: 41%

    Winning Trades:846 (69.63%), Losing Trades:369, 30.37%

    Avg($)*:43.492, AvgWin($):75.470, AvgLoss($):-29.825

    Avg(R)**:1.083, AvgWin(R):1.881, AvgLoss(R):-0.747

    * Assumes $2.5 of friction (including costs and slippage) per trade
    ** Measured in terms of unit of risk taken on each trade.

    Some more context:
    > Intraday - can be ~10min to all day
    > Not about chart patterns, i.e., doesn't look at 1/5/10 min bars etc. It looks for price movement characteristics in the context of recent price, broad market regime, etc.
    > Trades stocks that have 1M+ avg daily volume
    > Average ~200 shares per trade; obviously backtesting results scale w/ number of shares with I've kept trade size low.
    > Long and short
    > Avoids first 20-30 min to avoid high spreads/slippage. Avoids last 30 min.
    > There are probably ~2 direct input parameters but likely 5-7 parameters if I include stock selection, position sizing, etc. (cause for concern re: over-fitting)
    > ~1-2% account risk per position. Stop losses always in place. Conservative position sizing.
    > No margin (except for shorts obviously)

    I've tried several out of sample periods (including 2020 COVID correction/rally) and results are fairly consistent.

    Next I'm going to trade paper real-time through API . Obviously nothing beats putting this live in the market (especially since there are enough trades, i would know relatively soon before losing too much capital if this works)...BUT.... best to learn from others' experience.

    So... would appreciate any help/guidance on:
    a. Is the ~$43 avg trade too low for what costs + slippage might actually be? Recall, avoids first 20-30 min, vol >1M, 200 shares trade.
    b. Especially on shorts, is this likely?
    c. What else am I missing?

    This will not scale infinitely but I'm just looking to deploy some capital if it is robust.

    I am quite risk averse and skeptical so I am quite sure this will deteriorate significantly in real performance. Given the results, I'm worry about over-fitting a fair bit. I haven't traded intraday so new territory for me.

    Thoughts and guidance welcome.

    Thanks all.
     
    murray t turtle likes this.
  2. Regarding a) 200 shares of AMZN or F, big difference. I would suggest using other units.
     
    murray t turtle likes this.
  3. dholliday

    dholliday

    Interactive Brokers' paper-trading is quite realistic. I have run both paper and real, through the API, at the same time (stocks, intra-day, > 750K avg. vol.). Averages out to no significant difference. If it works, it works.
     
    murray t turtle likes this.
  4. SunTrader

    SunTrader

    No significant difference(?) - in the history of trading software testing that would be a first.
     
    murray t turtle likes this.
  5. bigmoetx

    bigmoetx

    Thanks. Agree. F vs. AMZN is a big difference.

    Just so I understand your push...here is more context: The system doesn't look to buy 200 shares outright but rather it looks to buy equal risk units (solving for no more than 1-2% of account equity).

    Over thousands of trades during backtest periods the average was actually closer to ~100 / share. But I recently increased the exposure/risk factors so now it would average closer to 200/share. Sometimes more.

    Let me know if that is aligned with what you were saying or if I should think about it differently.
     
  6. MarkBrown

    MarkBrown

    [​IMG]
     
    dholliday and murray t turtle like this.
  7. qlai

    qlai

    That’s my question- how will you (or your system) pick which stocks to trade?
     
  8. dholliday

    dholliday

    You can, of course, test this for yourself. They are not identical but they do average out. It's pretty clear that they put your paper-trade on the queue (or a parallel queue) and you get the price that comes up at your place in the queue. If I have to wait five minutes for a fill, it's five minutes on both accounts. As a side note; In the middle of a slow volume day, I have had to wait up to one-half hour to get a short filled.
    I've done this quite a bit over the years when a system first goes live or I make changes.
    I can't imagine trading a system that I don't first run on IBs paper-trading account. Never had a problem. If you can't make money on the paper-trading account, don't trade for real.
    Keep in mind that the demo account is not the same as the paper-trading account.
    I only trade stocks, intra-day, avg vol > 750K.
    All just my opinion. Wish all the best.
     
    murray t turtle and qlai like this.
  9. traider

    traider

    How much are you paying in comms?
    Are you able to negotiate better fees?
     
    murray t turtle likes this.
  10. Zwaen

    Zwaen

    I don't know the nature of your method but it is wise to also test a more gradually bearmarket (opposed to the 2020 correction). Especially if you use options volatility can have a far more different impact. And offcourse realistic spread-costs, if not mentioned already.
     
    #10     Feb 14, 2022