Discussion in 'Automated Trading' started by Zac, Nov 17, 2005.

  1. Zac


    I would appreciate any honest analysis of the following system results(all backtested).

    Its an EOD system, trading 18 ETFs. Upto a max of 4 positions can be held at one time, each position would be 50% of equity(using 2:1 margin)

    Annual Return 31.40 %
    Exposure 31.14 %
    Risk Adjusted Return % 100.86 %
    Winners (78.75 %)
    Avg. Profit % 2.92 %
    Max. Consecutive 20
    Avg. Loss % -3.39 %
    Max. Consecutive 6
    Max. system % drawdown -20.21 %
    Recovery Factor 7.63
    Profit Factor 3.09
    Payoff Ratio 0.83
    Risk-Reward Ratio 1.26
    Ulcer Index 5.09
    Ulcer Performance Index 5.11
    Sharpe Ratio of trades 2.19
    K-Ratio 0.07

    only honest feedback please!
  2. Can you provide definitions on how these factors were calculated? You also need to specify time frame, trade duration statistics, autocorrelation of returns, etc. Sharpe ratio is useless. Btw, daily data is a very dull instrument to measure prices with.

  3. Zac


    Thanks for your response.
    Heres the information you have asked for:
    Avg. Bars Held 9.91 <-- Daily
    Avg. Profit/Loss % 1.58 %
    Testing period is Jan 1997 to Jan 2005

    Ulcer Index &#8722; Square root of sum of squared drawdowns divided by number of bars
    Recovery Factor &#8722; Net profit divided by Max. system drawdown

    Profit Factor &#8722; Profit of winners divided by loss of losers

    Risk Adjusted Return % &#8722; Annual return % divided by Exposure %

    Max. system % drawdown &#8722; The largest peak to valley percentage decline experienced in portfolio

    Payoff Ratio &#8722; Ratio average win / average loss

    Ulcer Performance Index &#8722; (Annual profit &#8722; Tresury notes profit)/Ulcer Index'>Ulcer Performance
    Index. Currently tresury notes profit is hardcoded at 5.4.
  4. looks like wealth-lab output. the only thing is that ETFs are hard to borrow, and hence short.
  5. Seems reasonable.. I tend not to focus on all these types of stats so I'm not sure I can help there.. I don't even know what a bar is, I'm a programmer/automated trader/pseudo mathematicitian, not a trader.

    From a post on another forum which applies here.

    At a recent conference I once saw an intuitive and simple demonstration of the concept you were trying to get across. I thought to myself that some day i would need to use this to convince a less sophisticated person of the importance of autocorrelation of losses and of max drawdown. Maybe it was your talk I saw?

    The speaker showed two HF strategies: one strategy went up and down by about 2%, alternating, so returns were like -2%,+2%,-2%,+2% etc.

    The next strategy was one which over the same period of 30 days went down 1% 15 days in a row and then reversed itself back up to where it started over the remaining 15 days.

    Which was more risky? Well, the volatility of the first one, whose NAV varied between 98 and 100, was double the volatility of the second one. So it was more risky??

    Of course the second one had 15% peak to trough drawdown and nearly 100% autocorrelation of losses, so it was far worse.

    Probably a good estimate of real risk is to calculate autocorrelation of returns, estimate kurtis, skewness, also, you can estimate the tail-index to get an idea of the risk of extreme loss.

    However this can be very hard to do with a small sample size but there are some papers on how to calculate and maybe some matlab code laying around if you dig far enough.

  6. Is this an out-of-sample test, or the results of an optimization run?
    Did you include commission and slippage?
  7. Zac



    It is amibroker output. Out of the basket of ETFs only 6/7 trade less than 1 million per day. Shorting can be a problem for those. However good thing is only 8.3% of the trades (out of 310 in 8 years) were short. System shorts only when ETF is in bear market.
  8. Zac



    Data is from Jan 1997 to Jan 2005. System was developed for SPY, and same system was used for other ETfs, without any optimization. There were too few trades trading just SPY(approx 5-6 per year), however win rate was slightly better about 84%.

    YTD resulsts were 22% with 12% drawdown(peak-valley) and win rate if 77%.
    No commision/slippage is included. Using a $5 commission per trades decreases annual performance in back-test by 0.7%. System used closing price for trade, IMO doing a scan few mins before market close will even out the effect of slippage for popular ETFs e.g QQQQ, SPY, MDY, SPY, EWJ,XLE, however there can be slippage in other ETFs.
    will appreciate any feedback/criticism ..
  9. Zac


    Any feedback?
  10. The problem I would have is that you are leveraged to the max. And ETFs are highly correlated to the broader market, so you are explosing yourself big-time to the U.S. equities market, and for quite some time on many of your open positions. Some type of catastrophic market event could really clean your clock.

    The numbers may look solid to you. But the bigger question is, could you sleep at night? Can you open fairly long-term positions at full margin and just let them run?

    Easier said that done, believe me.
    #10     Nov 20, 2005