ForexKiwi's D1 Fibo trading auto-trader by Steve Hopwood

SERIOUS WARNING

1. Most Forex traders lose all their money.
2. Using the robot posted here in trading Forex does not guarantee success.
3. Trading this robot could lead to serious financial loss.
4. Trading this robot without understanding its underlying trading strategies guarantees traders will lose their money.
5. To trade this robot, you have to understand how to use the Fibonacci sequence in trading. You are unlikely to gain that knowledge here.

I awoke this morning determined not to code a single line. Sadly, I received a pm from ForexKiwi.


Hunker down guys, this is going to be a bit of a read.


Here is a digest of ForexKiwi's proposal:

1. The market on any given day is likely to follow the direction of the previous day.
2. ForexKiwi's research showed that when this happens, a 40 pip move is likely if the market is not trending strongly; a 100 pip move is likely in a strong trend.
3. If the previous candle was a riser, we are looking to buy. If it was a faller, we are looking to sell.
4. At the start of each new D1 candle, draw a fib from the high to the low of the previous day.
5. The key levels are the 75 (for buying) and 25 (for selling).
6. So:
   1. Rising candle:
      1. If the market is above the 75, send a pending buy limit pending trade.
      2. If the market is already below the 75, send an immediate market order

   2. Falling candle
      1. If the market is below the 25, send a pending sell limit pending trade.
      2. If the market is already above the 25, send an immediate market order

   3. TP and SL both 100

ForexKiwi asked if I can code the robot. What is a guy supposed to do?


So was born the unattractively-named 'ForexKiwi's D1 Fib Retracement auto-trader by SteveHopwood'. I am going to call it FK for short.


FK was an instant backtesting loser. We all know that I do not have a lot of faith in backtesting, but I can state this with certaintly; if an ea loses in backtesting, it gets murdered on a forward demo.


I was playing last night, and coded another Rsi trader to scalp EU. Go long if

1. D1 Rsi(20) is above 55 and is higher than it was yesterday
2. Chart TF Rsi(20) is higher than it was at the close of the previous candle

This was an instant backtesting loser, so I did some playing with optimising the tp and sl settings, and turned it into a quite magnificent winner. I am a newbie at optimising, so I may not be doing this correctly.


Applying this to FK, I optimised the tp and sl from 50 to 200 in steps of 10 over the last year and turned it into a winner. FK includes an option to use optimised values – AutoCalculateTpSl. The optimised values for the pairs I have optimised are hard-coded into init() and will be used to provide the tp and sl. You can read the list of pairs I have optimised further down; as I optimise more pairs, I will add to the list.


So, question no 1 for the experienced optimisers is this: how far back should I go in time when optimising the D1 chart?


The rest of FK is easy to understand. It draws the fib and sets the tp/sl when it sends trades at the start of each D1 candle. There are the usual management features. Pendings are deleted if unfilled after 18 hours (at least, that is what I intended to code).


Optimised pairs list

1. EURUSD
2. USDCHF
3. GBPUSD
4. USDJPY
5. GBPJPY
6. AUDUSD
   

Pairs and settings found most profitable by users

1. ForexKiwi: GPBUSD: http://www.forexfactory.com/showpost...6&postcount=33
   1. TP 300
   2. SL 130
   3. Breakeven 40: Breakeven pips 20
   4. Trailing Stop 30: Trail After Breakeven = False

The interesting stuff
Here is where things become complicated, so newbie traders and non-coders might like to skip the next bit.


I know I optimised back over a year, but I wonder if it is really the previous month that is more significant. This made me contemplate how great it would be if FK could do its own optimisation at the start of each new monthly candlestick.


Before I even begin to code self-optimisation, I need to design the algorithm. Here is what I am thinking. At the start of each new MN1 candle:

 

1. TP = 50, SL = 50, BiggestProfit = 0
2. open a TP loop
   1. open a SL loop with a LoopProfit = 0
      1. open a 30 D1 candle look-back loop, starting with the previous D1 candle
         1. read the high/low of the candle, define it as rising or falling, and calculate the 75/25 fib levels
         2. candle was rising:
            1. current candle opens above or at the 75: simulate a pending buy limit at the 75
            2. current candle opens below the 75: simulate a market buy order

         3. candle was falling:
            1. current candle opens below or at the 25: simulate a pending sell limit at the 25
            2. current candle opens above the 25: simulate a market sell order

         4. open a M1 loop indexed at the start of the D1 candlestick
            1. buy trade
               1. M1 high >= TP: add the TP pips to LoopProfit and close the loop
               2. M1 low <= SL: subtract the SL pips from LoopProfit and close the loop

            2. sell trade
               1. M1 low <= TP: add the TP pips to LoopProfit and close the loop
               2. M1 high >= SL: subtract the SL pips from LoopProfit and close the loop

            3. increment the loop index to move forward in time, closer to now

         5. end of M1 loop
         6. if LoopProfit > BiggestProfit then BiggestProfit becomes LoopProfit. The TP and SL that gave this figure are stored in a global variable or a file
         7. Increment the D1 candle index to go 1 day further back in time

      2. end of 30 D1 candle look-back loop
      3. Increment the SL by 10

   2. end of SL loop
   3. Increment the TP by 10

3. end of TP loop

At the end of this process, we have the TP and SL combo that gave the best returns over the previous month's trading, saved in global variables or a file that FK can re-read whenever it is restarted.


So, any further thoughts, guys? Anything wrong with the algorithm? I have attached the zip with the Open Office document I used to create this page, so you can re-order/correct/add to the algorithm should you see anything wrong.





http://cdn.forexfactory.com/images/s...m/big_grin.gif
___________________________________ ____________________________

Traders familiar with my work need read no further.

Posting restrictions
I have restricted the ability to post in this thread to: traders with at least 1 voucher; and traders on my 'buddy' list. I have done this because I do not want the thread cluttered with 'newbie' questions about the basics. New/inexperienced traders, this is not intended as a direct insult. What I want here is discussion with traders who fully understand what it is they are doing; when you newbs have been around a while, you will come to understand what a difference experience makes to the posts a member writes and shudder at the recollection of some of the questions you asked. I do. http://cdn.forexfactory.com/images/s..._the_floor.gif Go to your profile page; there is a link to the FF page that describes the voucher process.

Do not attempt to get around this by sending me pm's asking for help with your basic problems. Doing so will earn you an automatic ban from all of my threads.

I have set up a partner thread to this one and Nanningbob's at http://www.forexfactory.com/showthre...66#post3948666
Here, you can post the questions that you cannot post here, about anything you need to know. Someone with the knowledge and experience you need will answer. I opened it for Nanningbob traders, but feel free to ask about this robot as well.

Traders/coders with experience who have not yet been vouched for, can get around this restriction be being added to my 'buddy' list; all you have to do is pm me to ask me to do this.

Steve,

You are truly a self optimized machine; but take care, self optimization is full of trepidation.

Fringe NN (neurotic nuttiness) will test the boundaries of MT4 and before you know it, you will be contemplating a dll here or there.

A slippery slope my good friend. Consider the above.

Regards,

getdown71

Thanks Steve. Backtesting sure is a tad under-whelming.

I notice that it appears to be allowing multiple trades to be open at once rather than just allowing 1 trade at a time. The idea is that we only follow the previous days candle direction, unless it doesn't hit TP or SL - in which case we keep the trade open, but don't allow another trade to open until the previous one has closed. Not sure how much difference it will make in testing but it does seem to allow more losers than we had wanted.

Somehow I am trying to take advantage of the fact that if you look at a Daily chart there are often patterns of a few days in a row in the same direction and for a lot of pips - and even when it changes direction there is often a bit of a continuation for at least part of the next day, to get us to breakeven and limit our losers. The components of this strategy may need plenty of adapting to best take advantage of the above pattern.

I added code to stop multiple trading and it killed profitability stone dead. I have added a AllowMultipleTrades input so you can control this yourself.

Post 1.

Been testing this puppy and altering a few settings. If I changed the entry from the .75 and .25 Fib to immediate entry at 1.0 and 0.0 and change AllowMultipleTrades to False we get has a nice upward equity curve over the last year for GBPUSD with TP 300 and SL 130. EURUSD is also nicely profitable with TP 300 and SL 90.

Working on the other currencies to see what works for them.

My only reason for wanting AllowMultipleTrades to be False is to limit exposure and therefore be able to use a bit higher Lots. May not be the best, but we'll see. I plan to demo the above two currencies from tomorrow.

I should mention I also had Breakeven at 40 locking in 20 pips, and a Jumping Stop of 30.

Oh boy, but we have some big things coming.

I pm'd some coders I respect hugely, asking them to take a look at post 1. Squalou has responded by saying that he has already coded an auto-optimisation module that requires just some tweaking before making it available.

That will turn our trading robots into monsters.

One last one before I go out for my regular Sunday teaching.

You should see what sq has just sent me. Brain-blowing.

"some tweaking" might be just a little bit overexaggerated...
My optimizer uses the chart's bars to determine trades outcomes.
This EA being a D1 trader, the trade outcome cannot be determined with enough precision to be accurate.
It needs first some mods to allow it to run on M1 charts, so the optimizer code can be accurate.
The will also need somewhat heavy mods so that the MT4 trading functions like Ask, Bid, OrderSend(), OrderSelect(), etc, can be simulated properly...

It won't be as easy as we could think in the first place i'm affraid...


SQ

Sort of off topic but can this optimizer but included in the beast to optimize other currency pairs? I know we concluded it is best to just trade the majors but it doesn't hurt to leave a demo running on all the other pairs.

An optimiser could be fun to play with and make this method (and others)a lot more dynamic.

Regarding the Fib entry points my initial feeling had been that it was worth waiting for a fairly regular retrace back to the .618 Fib (with us taking it at .75 Fib to be safe), before it then carries on the direction from the day before. Backtesting however has suggested that on the strong move days we too often miss out on good trades because it doesn't retrace. I therefore believe that entering at between the .90 Fib and 1.0 Fib may be the better way to go. Will be interesting to see if optimising changes this at all.

Daily bars depend on the broker time shift, and 1 hour ahead or behind will probably give much different results.

In fact, rather than using D1 charts, we should rather see this system as a 1-trade-per-24-hour-period system, where the trade bell could be depending on the pair and the session (NY,Asian,London).

We should have a "trading hour" input for that.
And it could therefore be one the optimizable settings too, not only tp/sl.

If anyone is posting results here, at least he should give the gmt offset used for these results so we can reproduce them without walking in circles asking ourself "how the hell did he get those incredible results?" ...

SQ

Nice idea. My backtesting has been done on GMT +0.

Done. Post 1. The input is TradingHour. There is a resultant need for control over the expiry period for a pending trade, so this input is PendingExpiryHours.

These inputs work in your computer's local time. The defaults suit me here in the UK.

First day on Alpari UK demo. Interestingly it had a Buy on EURUSD and a Sell on GBPUSD and both worked. EURUSD carried on Fridays up move for 40 pips (20 locked in) before heading down, and GBPUSD Sell trade is still open and currently up 60 pips and accumulating.

Because it's a demo I also set it going on GBPJPY, EURJPY, USDCAD, and AUDUSD for fun. All actually are positive except GBPJPY which is down about 10 pips currently. Total trades +110 pips for the day currently.

Not a bad start.

beautyful!

I agree that optimizing off a D1 chart may be difficult. The problem is that its hard to look back far enough to generate quality statistics. Breaking down the last ten years and filing the data into days wouldn't be enough information to find patterns that aren't effected by news and can be predictably repeated. You need at least 250k bars.

I would suggest using a smaller time frame that still isn't too noisy.

Now, the other thing is:

as D1 candles will differ for different brokrs and different "trade bell" setting...
shouldn't our D1 bars be synchronized with "TradingHour"?

If we set TradingHour to 12, the brokr's D1 bars are late by 12 hours already.

Knowing that H1 bars will not differ for different broker shifts, we can build virtual D1 candles to be 24 consecutive H1 candles that start/end at TradingHour.

This would represent more accurately our actual D1 periods, right ?

The swinghigh/swinglow code could be something like this:

Then we need to take care of accounting for saturday/sunday to keep our 24 H1 candles...

And GetSwing() would be called every hour rather than every (D1) candle.

SQ

The interesting thing which couldn't be obvious during backtesting was the effective hedging that happened today. EURUSD, EURJPY and USDJPY were buys, while GBPUSD, GBPJPY, and AUDUSD were sells. Be interesting to see how often that happens. If it does happen a fair bit, the basket TP at a certain profit may be worth considering down track??

Steve and FK.......Steve another really nice Fib trader on a D1 chart.....FK a great strategy.......Looking forward to a lot of fun....

whip