Hi guys,

Wilko made an interesting comment in the library about Pathfinder and I really appreciate that kind of feedback. Wilko doubts that Pathfinder’s backtest result will be profitable in real trading.

@Wilko: It would be helpful for all of us if you could explain your arguments a little bit deeper. Do you made some tests or what is the basis of your opion? Do you have any ideas how we can improve or harden Pathfinder’s algorithm?

Thanks, Reiner

>>>>

Dear Reiner, dear fellow forum members,

After writing the previous post, I have myself decomposed the system into several independent systems and backtested them separately without the seasonailty boosting. What I found was that each of the parts/strategies independently generated not many trades over the full available database. In other words, allthough the outcome for each part/strategy may be positive, it is based on few observations, meaning the statistical expectation that historical results would repeat is weak. If you merely add a lot of weak parts/strategies together, this does not in itself guarntee the end-result is a strong (or rubust) system. Therefore, I would issue a warning: The parts/strategies incorporated in this system need to be tested independently on a much bigger dataset in order to be able to say with any statistical significance that the integrated system is likely to repeat past performance. In my book, one needs at least 100 observations for each of the different parts/strategies. To avoid misunderstanding, with parts I mean one part/strategy buys break of monthly high, the next part/strategy buys on break of weekly high, and so on. Pathfinder is comprised of 7 different parts/strategies. On top of that filters and seasonality boosts are applied, not uniformly, increasing the risk of curve-fitting. IMHO.

There are some nice ideas in the code that have been well applied, and for these ideas I thank Reiner!  I’m not saying the integrated system won’t work, I’m saying the certainty that it will work live is not high enough for me. I hope my findings can be of some help to fellow forum members. Please be careful out there! (Quote h/t the tv-series Hill Street Blues from the 80’s)

Miguel,

with the source code a good MT4 developer should be in a position to migrate Pathfinder to that platform. Describing 250 lines of code in words is’n easy 🙂 but the basic rules are:

• in the trading window (DAX 9-22) every H4 candle (for the DAX 9, 13, 17 and 2100 CET) the trading algo verify the 6 (variable l1-4 and s1-2) conditions

  // long position conditions
  l1 = signalline CROSSES OVER monthlyHigh
  l2 = signalline CROSSES OVER weeklyHigh
  l3 = signalline CROSSES OVER dailyHigh
  l4 = signalline CROSSES OVER monthlyLow
  
  // short position conditions
  s1 = signalline CROSSES UNDER monthlyHigh
  s2 = signalline CROSSES UNDER dailyLow

• for trading some conditions are combined with 4 Filters (one is your idea!)

  // define trading filters
  // 1. use fast and slow averages as filter because not every breakout is profitable
  f1 = close > Average[periodLongMA](close)
  f2 = close < Average[periodLongMA](close)
  f3 = close > Average[periodShortMA](close)
  
  // 2. check if position already reduced in trading window as additonal filter criteria
  alreadyReducedLongPosition = COUNTOFLONGSHARES  < startPositionLong
  alreadyReducedShortPosition = COUNTOFSHORTSHARES < startPositionShort
  
  // long entry with order cumulation
  IF ( (l1 OR l4 OR l2 OR (l3 AND f2)) AND NOT alreadyReducedLongPosition) THEN
  
  // short entry without order cumulation
  IF ( (s1 AND f3) OR (s2 AND f1) ) AND NOT alreadyReducedShortPosition THEN

• Key concept of Pathfinder is the smoothed average (if not available in MT4 Hull moving average delivers good results as well)

  // calculate instrument signalline with multiple smoothed averages
  firstMA = WilderAverage[periodFirstMA](close)
  secondMA = TimeSeriesAverage[periodSecondMA](firstMA)
  signalline = TimeSeriesAverage[periodThirdMA](secondMA)

• ProOrder manage the position in the following way
  • No pos: buy or sell
  • already long: cumulate in case of a long trade / sell the existing pos and go short in case of a short trade
  • already short: sell the existing long pos and go short in case of a short trade / do nothing because the system is already short

• The position size is boosted with a monthly saisonal multiplier until a maximal watermark
• money management will be managed sepratly for long and short trades in the following way:
  • overall take profit and stop loss (valid 24 hours)
  • trailling stop
  • maximal holding for profit and loss triggered by a maximal number of candles

remarks:

• a cross outside of the trading window will ignored
• sunday values will be considered for weekly and monthly high/low if available (DAX, FTSE, DOW)

I hoply consider all parts. Ask again, if something isn’t clear.

best, Reiner