Joe Doaks Backtests Unlikely Strategies

Discussion in 'Strategy Development' started by Joe Doaks, Jan 16, 2008.

  1. I recall fondly that when I began backtesting about eight years ago I discovered some really nifty systems. One of my favorites was "Buy Wednesday". It tested with tradeable positive expectation then, and I even traded it a bit. Cognoscenti of backtesting will perhaps understand why. In this thread I will revisit some of these early systems for the amusement of all. And perhaps I may even test some new systems for you. This is a simple system, as all truly profitable systems must be. I discovered over the years that if a system doesn't test well without profit targets or stops, it is self-deceptive to try to optimize it. The instrument is NQ. The period is all of 2007. Results include commission and slippage. The rule is simple: buy at the open every Wednesday, and sell at the close. It was a net loser by $1038, equity history attached for a typical poor-boy IB account of $5K.
  2. To answer the obvious question, "Sell Wednesday" was a net loser by $1151 in 2007. One of my favorite occupations years ago was deconstructing and simplifying SCT and testing the bits and pieces of it separately. That effort enormously enhanced my coding skill, but not, alas, my account balance. If anyone believes that any day of the week is a buy or sell, I will cheerfully demolish the notion upon a polite request.
  3. doli



    I have no doubts about your testing and trading abilities. I do wonder, however, whether a strategy without targets or stops is a buy-and-hold system. Do you ever exit?
  4. Thanks for asking, Doli. The BT's I will post are only for fun, or to poke fun. I do not BT seriously any more because I wasted a huge amount of time learning that optimizing profits and stops is a fool's errand for any practical system I ever found. Something as simple as an MA crossover system is hopelessly complex to optimize in the selection of the two MA's and for targets and stops. The system I scalp with has four rules and would be impossible to optimize with reasonable time and computing resources, even if you could trust the results not to be random, which you can't. But the next time a certain favorite poster of mine mouths off about buying at the close of an up day, I am going to nail him.
  5. Well here is a strategy which is psychologically satisfying to the dyspeptic hung-over trader: "Sell Monday"!

    Take it from me, if a system meanders between working and not working on the scale of a quarter, the results are random. Not something you will see in a textbook. Or did everybody really get bummed out every Monday in the fourth quarter? WTFK?

    Attached is the 2007 equity chart.
  6. spinoza


    Hi Joe,

    What happens if you are long EMD or ER2 only the last two trading days of each month and first day of the next month? Please test over past 10 years or so.

    That is, Enter Long near/at close of third to last trading day of month, and Exit Long position at/near close first trading day of next month. So, each trade lasts 3 days and overlaps end/beginning of each month.

    Thank you kind sir!
  7. Eddiefl



    Have you ever seen a daytrading system with E.M.A. crossovers worth anything.

    As surprised as I am I still see these advertised everywhere.

  8. Ah, my dear Bento, how I struggled in my youth to comprehend your writings! Perhaps I was hindered by being a heathen. No matter, we are all heathens on ET, filthy grubbers after lustrous lucre.

    I had not realized before that you are a comedian as well as a philosophe. I am but a poor boy who can afford only Easy-Signal as a data divider, and they most ass-suredly do not provide ten years of daily data. Alas, sadly, even if they did, I am an NQ fetishist and do not have a descryption to ER2 data.

    And worse, if I did, I would not offhand know how to code up finding the last two trading days of the month and the first. I am sure that I could eventail do it, but it is not worth the trouble to me en ce moment. If you have those aloggrisms I will reconsider. You might motorvate me to do so if you could explain why you think it might be a non-random systeme.
  9. Eddie, funny you should ask. Years ago I used the 15 and 75 SMAs on a one-minute chart to trade runs in NQ, based totally on eyeballing the data. Ah, the glory days, probably never to be repeated. Subsequently I wasted enormous effort to find crossunders (a new marketing term) that worked in a reduced volatility environment. To the best of my knowledge, there ain't any. The reason? Price action is now designed to spoof EVERY KNOWN TA method with a big one-minute price move which triggers trade signals on everyboody's charts. Which then of course are faded. I use two estimators which are roughly equivalent to EMAs as intraday S/R, but it will cost you a million in cash (euros, not dollaren) to find out what they are. All I will say is that the ratio of their speed is 4:1. They're not hard to find. But if you do, don't tell ET!
  10. Eddiefl


    LOL,, Joe, a valuable and entertaining response as i was expecting, you are like the Riddler in the Batman Series,, giving insightfull clues, but I respect that.

    This 4:1 speed ratio would be a stoch setting??, ,, please leave some crumbs, even the riddler left clues for the Batman,

    Thanks,, and I agree about the spoof and fading, it is a sick game they play.

    #10     Jan 17, 2008