Recent content by dom993

I haven't found that mythical Edge making someone profitable all at once, but I have found that profitability can come from adding a lot of little inches...

IMHO, 6 months out of sample for 18 months of training is not a validation at all. Your process has to overcome a huge datamining bias, of courses some strategies will pass your 6 months oos validation. I would suggest you to do exactly what you described using historical data from 2010-2011...

How will you validate anything coming out of this? What timeframe or combination of timeframes are you planning to look at? How much historical data do you have, of which quality? How long do you estimate the entire ML process will take, on how many compute engines? May be, you could try to...

The entry should always be at the open of the next bar ... I suppose your question is for MKT orders ... depending on the instrument, you can use the open with a couple ticks slippage (that would be the case for most liquid futures), but if you are looking at stocks, a MKT order placed before...

The OHLC of a bar are 4 samples of the transaction data stream. By the time you are sampling, the filtering of high frequencies should have already been done. It is a pb of bars creation that cannot be fixed after the fact.

In the backtest category, I forgot to mention, make sure commissions are realistic

It sounds like you have no experience at all with automated trading systems. On the other hand, I am one of NinjaTrader's partners, specialized in automated trading systems - which puts me in the vendor category, but I can also give you a few pointers. First, some systems are not tradable - by...

Also, is the system tradable live? How does it recover from a loss of connectivity, a local crash, etc ... backtest is one thing, live trading has a lot more issues

I am very skeptic about any backtest with Tradestation when they involve LMT orders (by default, these are deemed filled on a price touch instead of waiting for price to go through le limit). Also, if all you have to judge is 160 trades, backtest results are too easily overfit to mean anything...

An average isn't typical, and it is very misleading. Try plotting the distribution of the metrics you are interested in, and for that be sure to group your data points in buckets of equal size (eg is you are looking at an hourly range, use buckets in 10 pips increments : 1-10 / 11-20 / 21-30 /...

1. You should normalize everything to a standard position-size 2. Ignore the backtest & live stdev 3. Using the backtest trades distribution, run MC sim for the number of live trades you currently have, and take that stdev 4. Divide your normalized live avg/trade by the stdev just computed ...

With all due respect, the OP is telling us his backtest covers 49,000 trades over 40 pairs, and that's all the feedback you can provide? I challenge you to come with any system, overfit at will, showing positive expectancy over 49,000 or more trades, using conservative assumptions for...

Comparing the averages gives you nothing. Look how far your live average/trade is from the backtest one, not on $ value, but in number of stdev - of course, you must use for stdev what it would be for your live trades sample size, not the backtest sample size. To get that stdev value, a quick...

Very much in agreement with the above. I use NinjaTrader 7, on which I have developed automated systems which trade 24/5 with very minimal manual operations. I find C# the very best programming language hands down, it's been instrumental for me in developing the operations support layer of my...

Reason #1 : a lot of published backtest results come from unreliable backtest engines (Tradestation to be specific) ... when a LMT order is deemed filled on a price touch, the backtest results are simply garbage. Reason #2 : commissions and slippage are not accounted for.