So anyone who has been following this blog knows by now that I have not put a single dime of my real money into the Robotic portfolio yet, it's all been simulated trading and months now of trying to get a hold of some sense of reliability here...
I am a very detail oriented person, and from the very beginning things have not felt entirely right about this whole thing. From the way Scott communicated with me, to how long it took him to send me any info, which was just a slice of what I asked for, to the fact that half the info he sent me was totally bogus because he recorded it with errors on his chart settings, along how long it took us to figure out that was the problem (and him continually assuring me certain concerns I had were of no real concern), to TradeStation being an unstable shiny turd and not taking trades it should have taken, to them shutting down their Forex business, to getting the new MT4 robots that are missing key settings that the other robots actually utilized (Even though Scott has told me they do not utilize them)....
Well, since I couldn't freaking sleep again tonight I decided to take a closer look at the differences between the MT4 robot and the TradeStation robot. The MT4 robots haven't taken a single trade in over 24 hours, which is unusual for a portfolio of 12 robots I thought. So I opened up TradeStation to check and see if it had shown any trades for the portfolio as well. Thankfully there were no trades on that platform either, but I did get to thinking and checked some things out.
I backtested the USD/CHF pair in MT4, and it was profitable in the last year. I looked at the same pair in TradeStation, with the same settings and it had horribly lost money.
What the hell would cause that? Remember the post a little while back about how crazy the spreads are on some of these pairs? Well, that's always been a concern in the back of my mind: If these robots are taking as little as 5 pips profit, (And one of them did) it's very important that the testing include accurate slippage on each trade.
Well, I did some research on TradeStation settings and ran across two very important details that have changed things for me drastically. I found out that not only does TradeStation simulate some trades that will never take place in real life, (Which I'll explain in a moment) I also found out for certain that the only way to account for the spread is to enter the spread in the "Slippage" setting in the properties for the backtest.
I'd like to address the latter aspect first: the spread. Early on, I asked Scott about the spread being accurately represented in the TradeStation settings and the data in his spreadsheet. He has told me and others online continuously "I include enough slippage into the robot so that I actually do better in real life".
The bad news is that according to the live, real-time spreads shown on the actual MT4 platform in comparison to what is set in TradeStation, he is using half of the actual spread, and sometimes even less, which greatly effects the performance results as I will show you below.
The first point I mentioned, is that TradeStation will take a trade in the Strategy side of things at times, and will never actually execute it in real life and visa versa. Why? Because of how TradeStation tests. Listen Carefully and I'll try to make this as simple as I can.
MetaTrader tests with the spread included in the simulated price, therefore, if you set it for a 2 pip spread on the EUR/USD currency pair the buy price is 2 pips away from the sell price. 2 pips would be a price increment of 0.00020 just to be clear. In reality the price would look something like this:
Buy price: 1.11629 (also known as the ask, what sellers are asking for)
Sell price: 1.11609 (also known as the bid, what buyers are asking for)
However, TradeStation uses only the Sell price for both buying and selling in the strategy!
Buy price: 1.11609
Sell price: 1.11609
Why does that matter? I'll explain very clearly why.
Lets Say the market is dropping from 1.11720 and you have a buy order at 1.11609 because you think price will bottom out and turn around there. Because TradeStation backtesting only uses the Sell price to enter, if the sell price hits 1.11609 and price turns without going any lower, the strategy would think you took the trade, but in the real world you can't buy at the sell price. Because of the 2 pip spread, the real price you could buy at would be 1.11629 and that would mean you never actually took the trade.
Let's reverse the logic and say price was rising from 1.11520 up to a buy order you had at 1.11629. Lets say the sell price came up as far as 1.11609 and turned around. In real life, because of the 2 pip spread the buy price would have hit your entry at 1.11629 and your trade would have been triggered, when the strategy tester would have never entered the trade at all.
In both of these examples you can see a huge potential problem. In a currency where the spread is usually 1.5 pips, it may not be as likely that you will run into this problem. However, when the spread is 4 or 5 pips on a pair, this can happen even more depending on the volatility of the currency... It is a wider margin of error theoretically. This is a gigantic problem in my eyes and erodes the confidence needed to believe in this trading system based on TradeStation's method of testing.
I personally can't trust data that accounts for trades that would have never taken place or trades that would have happened in real life, that never show up in testing. Given the volatility of the market, I do think those occurrences would be somewhat rare, But who can really tell?
It may seem like no big deal, but who knows how many trades out of the thousands tested in TradeStation were BIG winners that would have never happened in real life thanks to a few measly pips not accounted for? And who can say there aren't missed losers for the same reason? You'd have to manually go back through the entire chart history and look at each trade to see if price turned around immediately. That would be exhaustive and even then, the robots are not the same for MT4 anyways. They are missing some trade settings. AND YES, THE ROBOTS ARE MISSING SETTINGS THAT THE TRADESTATION ROBOTS USED: An RSI exit that closes the position when overbought or oversold is reached. Here is just one example with the AUD/JPY.
So not only does TradeStation both potentially miss and add trades it shouldn't, it includes the slippage after the fact, which exacerbates the problem. Think of it like a ship in the sea. A small change in the sail at the beginning can produce a drastically different arrival point down the road. I imagine that may be what is happening here... The original TradeStation robot did so-so for 4 years and 9 months at 29%... But the MT4 versions adjusted version came in a half that amount! 15% gain in nearly 5 years is horrible!
That alone is a thorn in my mind, but given the other large problem being the fact that the spread was improperly input into the slippage... Once more accurately represented, it drastically alters the end result.
At this point, the only reliable testing out of the two for me, is MetaTrader, and those results have become unappealing with the more accurate spreads in the testing.
Perhaps a trend following portfolio, where the spread is of little consequence, is the way to go?
No comments:
Post a Comment