Back-Testing, Forward-Testing, Virtual-Testing (and finally make REAL MONEY!!) FOREX

[ozzymandius]

I've been reading these forums for a few months now and am about to explore Forex intraday trading. I am developing a number of simple, and not so simple, mechanical trading formulas that I want to test extensively before I do any real-money trading.

Fyi, I am not a programmer, but can handle basic technical tasks. I hope to find some intuitive software that does not require a computer science degree!

On that basis, I was hoping some of you more experienced traders could chime in with some software and other recommendations based on the following plan.


1. Purchase data disks of six years worth of intra-day historical data for major FOREX pairs.

2. Buy easy-to-program charting software like TradeStation or AmiBroker and input my mechanical trading formulas to create custom EAs.

3. Buy TradeSim backtesting software and import the six-years of historical data.

4. Export the custome EAs from TradeStation/AmiBroker to TradeSim and run hundreds (thousands? millions?) of Monte Carlo portfolio simulations to find and optimize the best EA over the 6 years of historical data.

5. Once the best EA is determined, manually "forward test" it using the charting software and six-years historical data by manually rolling forward through said data, and entering the buys / sells manually.

6. If the EA passes all of the above, then open a broker account and virtual trade ("paper trade") the EA in a semi-auto fashion (ie human hand needed to actually place buys/sells) for 30 days.

7. If EA performs as expected in virtual test, then open real-money account of only $100 to trade micros, again in semi-auto fashion for 30 days

8. If EA performs as expected with $100 account for 30 days, then repeat with $1,000 account,

9. If EA performs as expected with $1,000 account, then repeat with $25,000 account, still in semi-auto fashion.

10. If EA is still performing well after 6 months, then turn on FULL AUTO, but still spot check frequently, and halt trading if any "anomalous" results ever occur.


Needless to say, if the EA does not perform as expected in any of the above steps, it would be back to the drawing board for additional testing/tweaking. Also, please understand that I am not so unrealistic to expect that any system is going to be fool proof. What I am hoping to create is simply a mechanical system that makes a small, but steady profit over time - winning more trades than it loses.

So what I really want to know is:

a. Should I use Tradestation or Amibroker or another charting software? I need one that is easy to program and that will work with any of the major, reputable FOREX brokers. Metatrader's language looks too complex. The package must also be able to handle the manual "roll forward" of historical data outlined in step 5 above. And I must be able to create an EA that can one day be interfaced with the broker in a full auto mode.

b. Will TradeSim in conjunction with the charting software above allow me to back-test in the manner described in steps 3 and 4?

c. Are there any other software packages I should be considering?

Please also feel free to chime in on the basic steps of my plan. I think it is a sound and reasonable approach. But if you think it is stupid, naïve, off-base or lacking in any way - please feel free to say so and explain your reasoning. I learn best through criticism!

Thank you, and I hope to hear from any of you!

 

[ronblack]

5. Once the best EA is determined, manually "forward test" it using the charting software and six-years historical data by manually rolling forward through said data, and entering the buys / sells manually.

Why manually? There is no reason for that. it could take days. You can forward test the same way you backtest. It's one and the same thing.

Do not forget slippage and commissions.

Also, you sound to cautious about you trading $100 accounts. There is no need for wasting time to paper trade. Virtual trading has no connection to real trading. You'll be wasting your time.

Ron

 

[vladv]

Try to use a small account with REAL money,like 500-1000 USD.

 

[£10kLoser]

If back-testing really worked, don't you think one in all the millions that have been carried out would have come up with a 100% guaranteed winning system by now? What on earth makes you think that the market will behave this year like it did last year?

Nobody has ever proven that back testing has merit. Obviously and logically it can tell us something but back-testers curve fit their systems to work on the available data. You are then left with something that suffers from 'survivorship bias' - like we only see the fund managers that survive so they are bound to look reasonably good. The failures are eliminated giving us the illusion that what remains is better, when it is just luckier.

Most systems are the result of lucky back-testing matches and look good until the 'unforseen event' knocks them dead just like the systems that were rejected because they got knocked dead ealier. The sad fact is that with the numbers involved in trading (including the vast number of traders) unusually lucky runs are BOUND to exist and people see that as 'good trading'.

Runs of 10 blacks or reds in roulette are common now we can speed it up on the Internet. Likewise, runs of good and bad 'luck' plague trading and traders. Probabilty statistics and the understanding of probability is not something people have unless they have specialised in the subject.

Traders waste years and countless energy back-testing in the mad hope of finding a system when they should be learning how to trade...

 

[ronblack]

Traders waste years and countless energy back-testing in the mad hope of finding a system when they should be learning how to trade...

Excellent point. I agree that before back-testing systems traders should learn how to trade. That can help them understand the merits and limitations of back-testing.

Yep, it's true, very few understand probability theory and I suspect winning trader are part of those that do.

Ron