But what happens when we compare systems that have the exact same long term statistical characteristics but differ ONLY in trading frequency? If both systems have made the exact same amount of money with the same draw downs and other statistical characteristics but their only difference has been that one system has traded 100 times and the other 1000, which system is better ? In order to answer this question we need to understand the things that must be different in order to preserve system characteristics and have the same overall results.

First of all if system one (which will be our system that traded more) made the same money as system two then system two must have a much higher loss and profit target since it needs to lose and make as much money as system one on a much smaller number of trades. This implies that system two traded less positions with much wider targets to achieve the same statistical characteristics. Since the compounding of system two is also much less efficient it probably means that in order to attain the same profit it needs a more favorable risk to reward ratio since it needs a higher ratio to compensate for the effect of slower compounding.

The consequences of the above are very interesting. Since system two entered less trades with wider targets and a more favorable risk to reward ratio it is inherently less vulnerable to execution problems that system one. It means that any type of net slippage or spread variation will affect system one much more than system two since system one has inherent characteristics which make it better than two. The smaller trading costs from system two also means that its edge must be much smaller than that of system one since it needs to compensate less for these inherent expenses. As you see this implies that if you have two systems with the exact same characteristics then the system that trades LESS is a wiser choice than the system that trades more because you have less dependency on execution and the need for a smaller edge.

### 15 Responses to “Trade More or Trade Less: The Incidence of Trading Frequency in System Quality”

1. Marc says:

If you’ve tested the strategy properly (including slippage and commissions), all of the technical factors should be incorporated in your results. I would always trade the strategy with more trades, because the results will be more statistically significant. If one strategy has 10 trades per year and the other has 220, you would need to test the strategy with 10 trades for 22 years to get the same significance. Are you doing that? Why wouldn’t I just test the 220 trade strategy for 22 years and really understand what my profit potential/loss is?

I think the trade less strategy leads to more failed strategies. I could easily create a strategy over the last 22 days trading once per day with an unbelievable profit and I think you could too. That’s essentially what you have created by trading 10 times a year. That’s near the exact significance level. Either way, you should always check the significance of your backtest instead of thinking about it anecdotally:

In addition, can you specify your definition of drawdown. Standard Drawdown is the maximium loss from close to close. Since you are in trades for a much longer period, you really need to look at the trades to make a fair comparison. Many times with long term strategies, the market goes against you a very long way, but is hidden by a recovery in the performance report. You could go down $10K and come back and make a profit, but do you want to trade that way? It makes those strategies tough to trade, and you may blow up you account before it recovers. Smaller loss strategies allow you to guage that more easily. There aren’t many people who can hold on being down a few thousand dollars in a trade. That need’s to be a consideration. • admin says: Hi Marc, Thank you for your comment :o) You are largely ignoring several factors. The first is that you simply CANNOT include execution problems within your results since you cannot tell whether or not you would have been filled at a certain price or what slippage you might have received if you had not. Although you can take some considerations into account you simply cannot know for SURE what the results of trading that strategy in the past might have been with enough accuracy to eliminate all problem concerning live execution. Of course if you account for everything you are still neglecting a big problem in FOREX trading which pertains to broker dependency. The more trades you have within a strategy the more broker dependency issues will play a role. You’re also assuming that statistical significance spawns only from trade number which is wrong. A strategy that has 200 trades within the past 2 months is VERY different from a strategy that has 200 trades within the past 10 years. Although the statistical significance by only looking at trade number is the same you need to consider that the number of market conditions which one strategy eliminated in order to take those same number of trades was MUCH larger. It is much more statistically significant to be able to trade 200 times eliminating unprofitable conditions for 10 years than to do so for 2 months :o). It is very simplistic to look at things from a mere “trade number” perspective ! Regarding draw downs, I usually refer to draw down as the distance between an equity high and low which contains any such moves “against you” without trades being closed. In Asirikuy we have implemented several tools which allow us to take this into account in simulations and this is what I look at when I evaluate and compare systems. Certainly you’re right in that strategies that have largest temporary equity losses might be “tougher” to trade against some that take trades frequently but Monte Carlo simulations reveal that you would be talking about fairly the same thing. If one system you’re taking a given floating loss on the higher frequency trading system you’re taking X consecutive losses before recovering. It is not that one system is intrinsically easier to trade when compared to another as the measures that determine this are generally the pain an ulcer index which depend on draw down depth and length (not trade number). I hope this further exposes my arguments in favor of systems with less trade frequency regarding the accuracy of their evaluation, less predominance of execution issues and relevance of other factors. Thanks again for your comment, Best Regards, Daniel • Franco says: Hey Daniel, You are probably right but Marc’ comment makes more sense to me. The author of Evidence Based Technical Analysis is also clear on this issue. He states that the data mining bias is inversely proportional to the amount of trades executed. It also makes sense to me that for twice the amount of trades you need half the backtesting period for the same statistical significance. Market conditions also differ on different timeframes, and that is one issue that I have with long term systems at the moment. You will never find a system that has 5000+ trades with a profitable outcome (at least I have not been able to find one), so if a long term system has been traded for lets say 30 years and the amount of trades are in the thousands then won’t the system start to fail just like an intraday short term system starts to fail after an x amount of trades? • admin says: Hi Franco, Thank you for your comment :o) As I said on a reply to Marc’s comment you cannot simply judge statistical significance as a measure of the number of trades, this view is way too simplistic. Overall market conditions also do NOT change on smaller time frames as they do on large ones. Thinking that 200 bars on a one hour time frame represent the same changes of market conditions as 200 bars on a daily time frame is a BIG mistake. Overall market conditions change as a function of many different things but the main thing that seems to control what “condition” plays out the most is the change in daily volatility. With 200 bars in a one hour time frame you have almost no coverage of changes in daily volatility while with 200 bars on a daily time frame you get a good variation. This point is further shown by evaluation of systems across different time frames. I did a somewhat extensive study about this – which I may cover on a video or publish on CT – which shows how a walk forward analysis exemplifies this point and shows that systems need to be tested on significant changes of daily volatility REGARDLESS of how many trades they have. The point here is 200 hourly bars is NOT equal to 200 daily bars, statistical significance over varied market conditions is attained through withstanding periods of change of daily volatility. I hope this further explains my point :o) Best Regards, Daniel • Franco says: Hey Daniel, Thanks for reply :) This topic is more complex than I initially thought… Maybe a video will be a good idea. I would love to understand this completely as it directly affects just about every part of system optimization and design. • Marc says: Sure you can account for Broker dependency. Just take your average for a year for commissions, slippage and any other fees you pay. If you can’t get a true estimate of your costs, then how can you simulate any results? Furthermore, if your costs are that inconsistent, you should probably look at trading something else. You’re getting ripped off. If I’m wrong about the statistical significance, then what is right? The fundamental reason why you can gain an advantage trading is because there are people making decisions in their best interest at all times. The more times you prove that you understand the decision making process, the better off you are. It doesn’t matter what size chart you trade, the more instances that you can prove you method is successful, the better off you will be because you will be sure that your observation about the market has been consistent over time. Market conditions are the same on M5 and M1440. It just takes longer to play out. Thinking otherwise is the reason I believe that longer term strategies tend to lead to more system failures. I may be wrong, but here is what I was saying about the drawdown. We have two strategies with a max drawdown of$4000. Lets say I have a “High Frequency” strategy with a stoploss of $400. The maximum size that my account would have to be is$4399, because I know below that point my max drawdown would’ve increase. On the “stay the course” strategy, my stoploss is set at $1500 (if you even have one). That means that my account actually has to be$5499.

This is an example, but many of those longer term trend strategies have even bigger stops than that and it creates a tradability issue (especially for multiple contracts/highly levered Forex situations where this is magnified). Plus, for those who don’t fully understand it, it gives the appearance of a new “drawdown” even though the close hasn’t occurred.

Hi Marc,

Broker dependency is also very significant and important in Forex trading which is a very strong argument I have against going for a lower frequency equivalent if I have both systems that have the same exact characteristics but one of them trade less (although as I said on the article at least above some minimum trade number). You’re probably right in that this brings gigantic uncertainty to our trading and that ideally it would be better to trade another market (like futures where this doesn’t exist) but this effectively makes your statistical certainty start to decrease as a function of the problems of dependency as your system grows (at least in Forex trading). How would you take broker dependency into account when a) broker dependency follows no pattern b) broker dependency varies greatly amongst different brokers and c) brokers can change their feeds and liquidity providers at will ? Broker dependency cannot be taken into account by averaging because it affects each trade in a different way. If two systems have the same statistical characteristics (except for trade number) then there is no doubt that the lower frequency system gives less room to broker dependency.

Regarding market conditions and “playing out” I think you’re very wrong here. A system that takes 200 trades on the daily chart is not equivalent to a system that takes 200 trades on a 5M chart. View it in this way – forget about the time frame – the first system took its trades by analyzing and taking signals on an amount of data which was generated by X ticks while the second one used a tiny fraction of that amount. The first system uses a MUCH LARGER amount of information for its decisions and therefore its trading is more relevant than for the first. Limiting statistical relevance to trade number is too simplistic (as I said before).

Regarding draw down, I see what you mean now but I do not necessarily agree. If you evaluate draw down from equity high to equity low then if two systems have a 4000 USD draw down then they can both be stopped at the exact same time. The only difference would be that the second system would be stopped in the middle of an open position while the other will be stopped at an even number of closed positions. If however you need to have a closed position above your draw down then you’re absolutely right and the first system allows you to decide you’ve gone “above the limit” first. However if you look at the way in which we trade (where system usually never risk more than 0.5-1% of account equity) then waiting for an additional losing trade is not that bad.

The way in which we trade Forex at Asirikuy is also based on very careful risk management and all of our systems have stops which are usually set somewhere between 0.5-1% of the account’s size. Evidently systems – regardless of their frequency – need to be traded in such a way to make risks per trade reasonable and this is something we do not forget about when trading Forex systems. Again, when talking about frequency in Forex you need to consider that trading conditions are generally much harder than in any other market, we have broker dependency and execution issues which are a hard step to tackle using simulations, aspects which become more important at higher trading frequencies. If you would like to truly understand the way in which we trade and tackle the above issues then I would strongly encourage you to join Asirikuy :o) Thank you very much again for commenting,

Best Regards,

Daniel

• Marc says:

I guess I didn’t learn averaging properly in school. The way I learned it, so long as you have a large enough set of data, you can expect to go back to the average. The type of distribution doesn’t matter. Every distribution has a mean. Each individual instance doesn’t matter You’ll get 5 one time 2 another and 0 another. It’ll wash out. I know you are wrong here.

If your broker is so unreliable that you can’t get a true average for you costs, then something is grossly wrong and you should stop trading with them. The broker is so bad they’ve violated the rules of basic math.

In addition, having more trades actually helps the broker dependency issue. The more trades you have, the more likely you are to go to the mean. Few trades means that you may not have had enough trades to wash out broker dependency.

You’ve also confused the concept of statistical significance of a trading system. The amount of data taken to realize the entry conditions has nothing at all to do with the statistical significance of your trading system, so your example about using more information to make your decision is completely off base.

The statistical significance only has to do with how well your system will perform in the future and nothing about your actual trading methodology. That’s why I’ve been talking about number of trades. Its because the more trades you have in your test (especially out of sample), the more likely that situation is to play out in the future.

In any event, there is also no difference in the amount of information as you detailed. If you make a decision on a 14 bar indicator on a 5 minute chart and then make a decision on a 14 bar indicator on a 1 tick chart, then you’ve used the same amount of information. 14 bars. The only difference is the chart size. Does 70 minutes give you more information overall than 14 ticks? Yes. Did the amount of data you actually used in your decision making process change? No. Its 14 bars.

Hi Mark,

Thank you for your reply :o) Nothing better than a good discussion! Regarding the averaging, I propose you do the experiment so that you know what I’m talking about. Run a simulation with a given slippage on all trades and then run another in which the slippage is distributed according to some arbitrary random rule between 0 and a max value such that the outcome is the same as your average. You’ll see that since this not only affects costs but OUTCOMES of individual trades your net results will be different. Just do the experiment and you’ll see.

Again, it’s not that I can’t estimate my broker’s costs but that the data used for simulations is different than my brokers due to the inherent dependency between brokers in Forex trading. The data will be different and so will your results on a higher number of trades (bear in mind that we only have ONE 10 year data set). I have in fact observed that in some cases having more trades even helps you curve-fit to some particular aspect of a broker’s feed. Again, I have studied this deeply and I challenge you to do the experiments and see how this is the case and how the problem grows in fact worse as you increase trade number.

Regarding the concepts, perhaps you’re right here in the sense that we might be talking about different things. In my definition I take statistical significance as a measure of how the trades represent a statistical outcome outside of that which can be considered random. Since in forex trading a system that trades more and has the same characteristics as one that trades less has added problems with execution and broker dependency then this is in fact the case, the results have a smaller statistical significance because increasing the number of trades decreases your chance of distinguishing them from random chance. I agree however in that this may be a bell-shaped curve in the sense that it may increase significantly until a point were the issues dealing with execution start to become more important.

About the data used for the calculations, I think you are underestimating the importance of the information “bhind the numbers” you’re using. What is the difference between using 14 one minute bars and 14 daily bars ? That the daily bars are generated with a lot more of information than the 14 one minute bars. The amount of data (as the actual amount of numbers) is the same but the meaning behind them is dramatically different. It took 14 days of information to get one set and 14 minutes to get the other, these two cannot possibly yield systems with the same relevance. Again, I would encourage you to test this hypothesis and share your results with me.

As you know I am a guy who is all for the evidence and I have no intent of “imposing” a point of view so if you can offer conclusive points about your views regarding Forex trading and trade frequency I will be more than glad to incorporate them into my analysis and post an article about it. If you have this I would be thrilled to let you write a guest article if you want to :o) My article is the result of my personal research and experience but I will always, gladly take and incorporate points of view that are showed through conclusive evidence. Thank you very much again for your comments and for discussing matters in depth :o)

Best Regards,

Daniel

PS: You can send me an email to dfernandezp (..at..) unal.edu.co if you want to share any test results you have on the matter

2. John says:

As I browse the FX trading forums, one thing I notice is how many traders are impatient and think all systems need to trade multipe times per day. I see many comments for EAs like “This thing hasn’t traded in 36 hours…is it dead?” or “This EA only traded 4 times in the last 3 weeks, I’m moving on to something else.”

That’s one of many reasons that patient, realistic traders who trade long-term profitable systems have an edge over the rest…

• Franco says:

That is true, I have a lot of respect for manual traders, don’t know how they do it having to sit all day watching the charts without taking action will kill me.

Maybe that is the secret to trading ;)

3. andrea says:

Hi Daniel,
I think that when I trade a system on daily chart, the slippage and spread aren’t so important than on 4H, 1H or less.
So I prefer a system that trade more often. What do you prefer: you have 2 sytem with the same final result (profit and drawdown)a system that makes 20 trade in 10 year (on daily time frame), with 2 or 3 trade that makes all the profit, or a system that makes 200 trades (on daily time frame) with 50 trades tha makes the profit?
Andrea

Hi Andrea,

Thank you for your comment :o) Well when you put it at such extremes the answer seems to be obvious (the second system) because the first one has simply so few trades. However it becomes more significant once you have a larger number of trades, for example would you trade a system with 500 or with 1000 trades?
what about 1000 against 10,000 ? I guess there are many things yet to say about this issue and a second post might be worth it! Thanks again for posting,

Best Regards,

Daniel

4. Vitor says:

I know experienced manual trader that says: experience traders, trade systems with bigger time frames with less frequency trades and newbies tend to use systems like scapling with smaller time frames.
He always let “nearly perfect” trade set up formed pass by and only trade very few times on the week. He says, you dont have to worry to let a "nearly perfect" trade set up pass, the forex is full of oportunities, tomorrow there is another one, trade only when you its the perfect time(according to your system plan).

5. alex says:

if you have 2 systems which differ only in number of trades you have to be a little stupid not to chose the one that has the highest number of trades. the profit formula is very clear. if you have the same system parameters but different number of trades the most profitable one will be the one with the highest number of trades.