Thanks! The odd lot numbers come straight out of my models, and it wouldn’t be a problem to start using round-lots. I think, it must be either the odd lots or volume (which is for those four stocks between 0.2 and 1% of total volume for the stock on the day of the order, except for one order). Does someone know which of the two has the greatest effect on the realism factor?
Francis … this is exactly the issue I want to test out next. I have always used the TB setting to convert limit orders to market orders within X seconds, and played around with X (settled on X=2 to get the job done as quickly as possible). It was more important to me to stay in synch autotrading than to miss a trade and get out of synch, so I wanted to get filled on each trade at all costs. The net result has been the 70% factor.
Since I trade multiple systems, using the limit order offset is not as obvious for one system as it might be for another. For example, with ATDow I’ve never missed a trade but can encounter huge slippage (eg. 13 YM points on a trade last week) in a fast market when the limit price is just touched. For this system I think I would be much better off using a 1 tick offset on the limit order, on average, and taking the worse fills. The 30% profit loss I am realizing with that system is due entirely to slippage … I’d guess that about 60% of the fills I get are after conversion to a market order by TB rather than hitting the limit initially (based on my fill prices, and I can quantity this if anyone is interested since I keep track of every trade and fill). That is, most orders are filled (both entry and exit) after being converted to market orders once C2 reports a fill rather from hitting the limit real time.
Another system I trade is WhyNot which trades much slower, but one tick in the US/ZB contract on 19 contracts (the current trade size) is $594. SInce the trades have entry/exit on stop and limit orders, and the trading time frame is slow (hours to days) there is much less of a problem with fast markets and just touched limit orders. There I’d hate to give away 1 tick on each side of the trade since the chance of a fast market event is much less likely, and the ticks are much more expensive ($594/tick on 19 US/ZB contracts vs. $45/tick on 9 YM contracts)…
Since TB doesn’t allow separate settings for each C2 system it not as easy to fiddle with the settings to try things out. But I think for ATDow I’ could improve on the 70% by using a 1 tick offset (possibly 2 ticks) on the system limit orders and eliminate the convert to market after X seconds once C2 reports a fill.
Francis,
I’m not Randy but I can see the logic in his reasoning, if this happens when (in case of a BTO order) the price is going up. The conversion to market if C2 reports a fill will then lead to a worse fill than on C2. One reason why I use my settings is because my impression is that extreme-os is often buying stocks when they are decreasing. In that case, more delay would give a better price. However, these are only impressions. I agree with Randy that the only way to determine the optimal setting is to have at least two test accounts.
Jules
This is why I was surprised at your negative limit order offset setting initially. As you say, Extreme-OS is a long-only system that tries to buy at a low price, and my missed fills there were always due to the limit price being just touched and the market turning around quickly. Since my main goal was to get 100% fills to stay in synch with C2 (rather than trying to get better fills in the cases when the market ran sufficiently through the limit) I would want a positive offset to guarantee fills.
In terms of net “RF”, your case of losing less on the losing trades (due to the better fills from the negative offset) and taking market fills on missed limits (which can go either way) may well work out better than my using a positive limit offset sufficient to get 100% fills, and no further fill management. I don’t want to resubscribe to the system (or any system) to try this out with real dollars, but it isn’t obvious which approach would work the best. I expect it will vary with the system to produce maximum complication!
"I expect it will vary with the system to produce maximum complication! "
I bet it does! 
Moreover, discussions like this may complicate matters even further, if many subscribers of extreme-os are going to copy my settings, thus ruining my market orders…
But thinking about this a little longer while I walked my dog, I’m not sure that I was right in identifying this as a property of extreme-os. Is it not more generally a property of limit orders that they probably will get filled (in case of a BTO) when the price is decreasing? So I would think that a negative offset and conversion delay might work for systems that send their orders some time ahead, trying to fish at the bottom / catch a falling knife. A positive offset would be more logical for trend following systems that buy when the price is already increasing.
PS Science trader, I was a little sloppy: The lower part of my earlier response to Randy, with the parts of the log file, was meant as an answer to your question of how to increase your RF.
Science trader,
You asked which one (odd lots or volume) has the greatests effect on the Realism Factor. If you go to your systems page, to the trade details, you can click “Realism Factor (RF) Calcs” on. This will display the details of the RF calculations for each trade separately, if you go with the mouse over the right area of the trade. As Matthew explained, the overall RF for your system is a weighted average of these tradewise RFs. I’m not a subscriber, so I can’t see all trades, but in all trades that I see the RF of the trade is equal to the Volume factor, and there are no slippage data available. I don’t know what the Volume factor is, but this suggests that the volumes are the most important in your RF. I’m not sure how you can change that. I expect that market orders are the only way to get a high RF in this case.
Jules
Jules, thanks for your help. I looked at these things myself, and also came to the conclusion that it’s the volume that must be driving these things. But consider my EXM trade, with a realism factor of ~62%:
- Both the buy and the sell order were at market, not limited.
- The total volume traded for EXM on both days was ~450 times the volume of my order.
That seems 100% realistic to me, except if there would be a substantial penalty for odd-lot orders.
Jules … the problem with any limit order srder system (buy side or sell side) is when the market just touches the limit price and turns around. If the price goes through the limit by one tick on either type of order everyone should get filled provided the volume is there. If the price misses the limit by one or more ticks, no one gets filled.
Francis can confirm this, but I would assume the sign of the limit order offset in TB is adjusted depending on whether the limit order is a buy order or a sell order. That is, TB will always add the offset to a buy order and subtract it from a sell order.
With a negative offset you decrease the chance of getting filled on these just-touched events, but get better prices if you are filled. So the negative offset may have some advantages, but I’d bet you a Euro TB added this feature assuming positive offsets, with the sign adjusted automatically depending on whether you have a buy or sell order.
Science trader,
So your market orders got a RF < 100. Interesting. I’'m not sure whether I should defend that, but from a subscribers point of view this could make sense: A subscriber wants to know whether he would get the same result as you. Now suppose you give a BTO market order when the ask is 15.00 but the volume of that ask (not the total volume of that day) is small, then C2 should fill you at 15.00 (best case) but it should also communicate that subscribers will probably get another fill (RF < 100).
However, what I remember from other discussions is that the RF in these cases is not determined by the volume of the ask, but the volume of the actual transactions. (I’m not entirely sure of this, see a thread about the RF a few weeks ago). It seems near to impossible to get a RF = 100 in this case.
Given what Matthew explained, it could help if you have autotrading subscribers that produce a ‘statistically significant’ set of slippage data. This would bring you in his case (A). Perhaps you can give yourself and some friends a free trial? Well, that must be pretty good friends then
Jules
Randy, I agree, and I don’t take that bet… I was probably as surprised as you were that TB accepted this and interpreted it in the way I wanted.
I understand your point, and agree that C2 should communicate that subscribers will probably get another fill if that is likely. However, to what extend "another fill" leads to unrealistic overall results, seems a different matter. If there are few trades, with large profits or losses per trade, the effect of a bad fill is less severe than in a high-frequency system. If you average over the realism factors per trade, this property gets lost.
Take for example an instrument that gains 20% in 10 days. Suppose that the loss due to slippage per trade is high, ~2% on average on both sides–each trade (ordered at market) would result in a low but similar realism factor. If I would buy and hold the instrument for the 10 days, net profit would be ~16%. If I would buy the stock at the beginning, and subsequently sell it and buy it back immediately for 4 times during those 10 days, and then sell it at the end of the 10 days, net profit would be < 0%.
In both scenarios the overall realism factor and unadjusted profit would be the same, even though in the first system I would come much closer to the theoretical profit of 20%.
> I think, it must be either the odd lots or volume (which is for those four stocks between 0.2 and 1% of total volume for the stock on the day of the order, except for one order)…
It really is about volume at price (in real life), not volume for the day. For
sure the odd lots would get filled in pieces.
"Does someone know which of the two has the greatest effect on the realism factor?"
OTOH, perhaps you should strive for the worst possible RF. As Matthew
points out the worse the RF, the better your adjusted results are…
Yes TB will add the offset if it’s a buy and substract it if it’s a sell.
Francis
> If you average over the realism factors per trade, this property gets lost.
Maybe, maybe not. There was a system that "bought" 1.5 million shares
(or something like that) in a stock with .4 million in total daily volume. The trade made several $100K in profits. 90%+ of it was fantasy and it accounted for most of the systems total profit of several $100K.
If I understand Matthew correctly, then in case (A) the RF is estimated for each trade separately, and the adjustments are also done per trade. Then the adjusted net profit of the low frequency system would be 16%, and the adjusted net profit of the high frequency system would be < 0%, as you describe. So I don’t think that there is a problem there.
But what worries me is that it seems that in case (B) the effect of missing trades and the effect of slippage seem to be confused. For profitable trades they would indeed have a similar effect (decreasing the profit), but for losing trades the effect of missing the trade is that the mean loss is smaller while the effect of slippage is probably that the mean loss is larger.
Were all your orders market orders? IF that is the case, then no subscriber would have missed the trades, but most of them would have had some slippage in comparison to the C2 results. Then the adjusted curve should be below the unadjusted curve, not above it.
Moreover, I don’t see how the volume at a certain price can indicate something about the slippage. OK, a small volume indicates that there would be slippage, but it doesn’t indicate something about the size of the slippage. I think that it would be better to base an estimate of the slippage on the depth of the orderbook or, if that is not available, on the price after an average delay. But then I think that this has been discussed before, and the only thing that I remember from it is that it was complicated, so probably I should shut up before I’ve to withdraw my comments again.
I hope this makes sense. I’m a little drunk as we lost the match against Portugal. (If we had won I would also be drunk, but then I wouldn’t be sitting here ;-))
Jules
I’m currently most concerned with what Matthew calls “case B”. And especially the following: "While C2 does indeed assign an overall Realism Factor (RF) to a system, that RF score is actually a weighted average of per-trade Realism Factors."
If you would simply average over the per-trade RF’s to obtain the overall RF score, my 2 scenarios seem to produce identical overall RF’s (not talking about the curves here, but about the single score).
To answer your question, most of the orders for my open positions were limit orders, although I did some at market, to see if there would be any effect on my RF. It is too bad that I can’t see the RF-per-trade for my open trades.
Ah, I understand your point. Yes, they would lead to identical overall RF. Whether that is appropriate, depends on whether the trade RFs represented missing trades or slippage. If they were caused by missing trades, then it is appropriate, I believe (averaged over subscribers, missing the long trade with 50% probability would have the same expected value as missing each of the short trades with 50% probability). If the trade RFs represent slippage (like in a market order with limited volume), then it would not be correct and the high frequency system should obtain a lower overall RF than the low frequency system.
Jules
what is the level if slippage for 100000 $ you use for each kind of products : stocks, futures, forex… ?
http://www.equitydream.com
Franklin, I’m not sure if I understand your question. Can you be a little more specific? FYI, I only trade stocks, not futures or forex, with this system.
