Trading Systems: A Valuation Based Approach

[Predictor 11]

One of my systems has historically returned, via hypothetical backtest, around 60%-80% per year (for the past 10 years) when tested with a wide stop loss. The return jumps into the 100%-120% annual range when the system forgoes the stop completely. A 10k account will grow to over 2.5 million dollars in just 8 years with a 100% annual compounded return!

 

These results lead to some interesting questions. First, we always read that past performance is not indicative of future performance. Obviously, various agencies and regulatory authorities require such disclaimers. Yet, anyone who trades a system believes that past performance is a good indication of future returns. I doubt there is any trader who doesn’t believe that. Of course, traders understand that such extrapolations are often wrong. Yet, I don’t know anyone who would trade if they didn’t on some level believe that historical performance had value. Even if its not a backtest and someone claims that the performance in a future test is important, the implication is the same which is that eventually your past performance has some value.

 

A reasonable question is that if we rely on the historical results to guide our trading and the system performs better without a stop then what is the justification for using a stop loss. Personally, I do like the idea of the stop loss and I understand that one bad trade could wipe out a highly leveraged system as I describe. In fact, the variation that I offer has such a stop loss and stop loss guidance for this very reason. One reason that’s the case is that its my personal preference, secondly the system with the stop still produces a strong return, and last but not least is that I do design systems, like any product, for a target audience. I know that my consumers would not accept a system without a stop loss. Yet, all those reasons aside there is a strong logic that if we accept the historical results as being indicative of future returns, which we already have by virtue of using the historical returns in the first place, then without a rational basis we should trade the best returning system -- which in this case was the system that didn’t use any stop loss!

 

I know that some will rush to judgment and assume that trading with very high leverage and no stop loss that one would be destined to blow up the account. However, the system I’m describing is a real system and was backtest over 10 to 12 years and the largest loss without a stop would have been around $2500 and there were probably no more then a couple losses that large out of hundreds of trades.

 

It could also make sense to use a stop loss wider then any historical loss. Such a stop loss would both be unlikely to ever be hit and could provide much needed peace of mind and would, moreover, protect the majority of the account if a truly unusual deviation were to occur.

 

I think part of the reason we like stop losses is because we don’t really want to accept the risk inherent in trading. Perhaps more to the point, we believe we can time the system or we’ll get lucky and avoid the historical losses. As a discretionary trader, I do feel that it is possible sometimes to time systems. Yet, if we choose to rely on a statistical methodology, that is not introducing unknowns, then it is rather likely that the account will be completely exhausted or mostly exhausted by the time we could recognize that the system had failed.

 

A different approach would be to consciously allocate the entire starting balance to the system with the intention of trading the system until failure or success. It is basically a bet the farm approach. The mindset has some advantages. One primary advantage is that such an approach is simply the act of consciously realizing and accepting the risk inherent in trying to achieve the desired returns. Following this logic, we would also see that using the stop loss wasn’t optimal and may choose rationally to trade without the stop with the full knowledge that single trade could wipe out the entire account. Mathematically, sense we are already extrapolating the past performance into the future, we can also extrapolate the returns and come up with a “valuation” for the system based on the historical performance. A system that returned 100% with a 10k starting balance would be worth, as described, approximately 2 million dollars with an 8 year horizon. We could even look at it like a single trade. In this case, it would represent a 200x potential return for the starting risk!

 

There is a problem with the extrapolated returns though. Even accepting that it could make sense to risk a small nest egg on a high risk trading system: it is much harder to justify the risk required to meet the returns extrapolated. In other words, one would have to risk 100k to make 200k and 200k to make 500k and 500k to get to a million and a million to get to 2 million. At some level, the risk required to make more just simply doesn’t make sense. As such, the extrapolated return isn’t realistic and therefore we face that problem of valuation again. Remember, we had justified trading our 10k starting capital based on the valuation that the system would produce over 2 million dollars.

Clearly, this thought experiment is missing a very critical element required for implementation. The missing element was first introduced to me from another trading associate and CTA, and that missing element is the systematic method of taking out profits as the account balance grows in value. This cash flow principle completely solves the problem conceptually and completes this idea. The extrapolated returns will, of course, be reduced but the overall theses becomes workable. In fact, there is certainly a possibility that the system could fail and take a complete account loss while the trader actually produced a net profit. If the system generates enough cash flow to “pay off” the initial investment before failure then one can walk away with a net profit even in the event of system failure. The method for this calculation is based on fractional position sizing and is outside the scope of this article.

 

Completing the paradigm would involve trading multiple systems and treating each system like an individual trade. The stop loss would be the starting account value and each “trade” or system would offer a very high extrapolated return compared to the starting balance. In this paradigm, a total account loss doesn’t infer that one “failed” as a trader but merely the system failed. The reward to the risk taken was already accepted. It goes without saying that I would only even attempt this with systems that I had a very high confidence in because bad systems could easily result in a total loss. But, let's assume we start with 50k and 5 solid systems with 100% annual returns per system and an 8 year time horizon. We'll reduce the implied return to 1 million per system due to the anticipated withdrawals. The actual reduction depends on the rate that we withdrawal profits. Furthermore let's assume that every single system but a single one fails. The net result is still that we made 1 million dollars from a 50k starting investment and that doesn't even take into the very real possibility that some systems could have payed off the initial investment and even made some profits before failure! The net result is achieving nearly 50% annualized return with an 80% system failure rate!

 

It is clear that having a plan to withdrawal and protect profits is the cornerstone that completes such an ambitious plan. The trader who wishes to execute such an ambitious methodology needs both the capital for at least a few seed systems and perhaps more importantly the wherewithal to execute the plan over many years. Notably, there is no requirement that the trader/system developer remain inactive and dormant over so many years. The more professional approach would be to invest some of the cash flow into new systems, update working systems, and adjust allocations based on performance and market conditions.

 

--

Curtis

http://themarketpredictor.com

[ratmpower 10]

Nice post. Like many traders, I also back test my strategies and believe that past performance is somewhat indicative of future performance.

[tommaso 10]

  Predictor said:

One of my systems has historically returned, via hypothetical backtest, around 60%-80% per year (for the past 10 years) when tested with a wide stop loss. The return jumps into the 100%-120% annual range when the system forgoes the stop completely. A 10k account will grow to over 2.5 million dollars in just 8 years with a 100% annual compounded return!

 

-

Curtis

The Market Predictor

 

Using the 1 realization of past data will only lead to curve fitting.

 

"Prediction" makes no sense as far as trading is concerned.

 

People easily forget their observations are submerged in an ocean of variance...

 

Tom

[SIUYA 11]

a good bunch of questions raised there predictor and as you say it will cause many varying responses of "this is right, this is wrong", which for I dont think your article deserves as its an interesting thought piece.....so as more food for thought.

 

One big difference between varying tests is in what you are trying to achieve and for whom you are trying to do it. For many they test to be able to grow and compound returns - for clients, while as an individual trader you might have far more degrees of freedom in this respect. Also maybe I missed it, but there are big liquidity issues with the compounding around many tests - depending on sizes of accounts....not such an issue for small accounts of course. Clients dont pay you to not grow the account....one issue is that if you are too successful they take money, if you under perform they take money, if you sit on cash, they pull money when you are at your worst backtested expected drawdown just as its about to turn around etc;....

 

On the issue with stops.....you must have some sort of take profit, or exit level, or you could easily get into a position at the incorrect time and only do one trade, never to see the instrument trigger again....and this raises another issue with some testing - it can be argued that you can never do enough testing - is 10 years of data enough - 30, 50 years. Many systems work in certain markets and for some we have had bull markets for the last 30 years.

 

diversifying systems and instruments it could be argued is the last real free edge available?? and it is precisely this that allows the compounding of returns.....

eg; 10 instruments/strategies,

year 1, 9 breakeven one makes 100% - funds under mgmt grow by 10%

year 2, different 9 BE, one makes 100% - funds under mgmt grow by 10% from a base of 110%...

etc; etc;

 

IMHO...backtesting gives nothing more than a road map of what to expect in terms of return and cashflow movement assuming that the market conditions you tested in remain the same going forward.

So I guess the big difference is that the assumption you make is not that the backtesting predicts future returns, but rather that the backtesting offers an insight into your predictions of future market behaviour.(you may be wrong) Hence another reason to have various non correlated (this is the crucial element) systems.

 

Or alternatively as you suggest - just go for it and realise that luck in terms of timing and not skill is what pays.

[joshdance 10]

  Predictor said:

I think part of the reason we like stop losses is because we don’t really want to accept the risk inherent in trading.

 

I believe just the opposite to be true: a person who uses a stop loss has accepted the financial risk, and quantified it. Someone who does not use a stop loss has not accepted the possibility that he is wrong if the instrument trades to some point. Therefore, the person who has no stop loss has not really accepted any risk at all, because it's not really possible for him to lose.

[YertleTurtle 10]

Trading without a stop loss is disregarding the number one rule in trading. Protect your capital! It makes me think of all the investors out there that buy a stock and watch it plummet but never think to sell. They made their buy and hold decision and they will hold regardless.

 

Its a terrible idea to trade without stops. You no longer have an exit strategy and emotions will be making your decisions. When someone asks what your risk reward ratio is you can proudly tell them anywhere from 0 to 100 to 1.

[eqsys 10]

There are systems that use stops, then there are systems that use rules to exit or enter trades. For example, with a stop system you are long something at 100 and have a trailing stop at, let's say, 90. My experience has been that this stop at 90 is highly likely to be hit on some brief selloff - and you are filled at some price BELOW 90. Then, frequently, the market reverses and you are sorry you placed the stop in the first place.

 

The second approach is based on rules. For example, if X and Y and Z then sell. This still gets you out of a bad trade, but tends to avoid some whipsaws. I have found that this type of system produces better results than just using a raw stop.

 

On backtesting: obviously a backtest is the best that can be achieved on a given set of data that was generated by a particular type of market condition - trending or non-trending. If you come up with a system based on a trending market, and the market stops trending, your system will lose money going forward, as the market no longer reflects the conditions that were in existence when you ran your backtest. Markets switch between trending and chopping all the time.

 

On drawdown: it's the drawdown that gets you in the end. If you are trading a $10k account and have a worst trade of $2.5k and a drawdown of $4k, you might still be able to sleep at night. However, if you are trading a $100k account and have $40k drawdown, THAT might get your attention. When your software calculates a theoretical backtest result, it doesn't have any emotion during the periods of high drawdown. However, YOU, as a rational human being, are not a computer, and when you are trading real money, the drawdown takes on a whole different meaning.

 

I have found, in real time trading, that your max drawdown will happen again, and often frequently - so don't assume that this is an "outlier" event. If you are not prepared, both financially and emotionally, for max drawdown, when it happens that will be end of your trading.

 

Best of luck and trading...

[SunTrader 10]

A valuation based approach, did I miss something somewhere such as ...... ummm valuation.

 

Where?

 

Stops / no stops is not a valuation approach if you ask me.

[Predictor 11]

The valuation comes from extrapolating the returns over the system some years into the future. The basic idea is that if I start to trade a system then I've already extrapolated the results into the future and therefore I can extrapolate the returns into the future. The real key to making this work though is that I take out my profits as the account increases in value at a defined rate. Based on the historical return rate and my cash flow plan then I can come up with a realistic valuation that doesn't require for me to risk "all future gains".

 

  SunTrader said:

A valuation based approach, did I miss something somewhere such as ...... ummm valuation.

 

Where?

 

Stops / no stops is not a valuation approach if you ask me.

[Predictor 11]

This is a reasonable perspective. However, we are referring to a system in this case which had a historical drawdown and return. I may have should been clear that there are exit rules in a system like this. It takes losses: it just doesn't take stop losses. As I noted, its not about wanting to use a stop or not use a stop. In this case, the system returned more without using the stop.

 

I should probably add too and this maybe something new for those who haven't built system is that typically the tighter the stop one uses on a system then the GREATER the drawdown. Let me repeat, the tighter the stop then the systems will typically experience GREATER drawdown. Of course, there is value in decreasing the maximum risk per trade because with less max risk per trade it is possible to leverage the system higher.

 

My perspective in this article was that a person starts to trade a system that historically had DD of 40% or 50% because they know they will only have to risk x$ because the system uses a stop. However, they are really kidding themselves versus accepting the risk of taking that 40% or 50% loss.

 

Where I agree with you.. is that yes if a trader doesn't have a plan to take losses and doesn't have an edge then that's a disaster waiting to happen.

 

  joshdance said:

I believe just the opposite to be true: a person who uses a stop loss has accepted the financial risk, and quantified it. Someone who does not use a stop loss has not accepted the possibility that he is wrong if the instrument trades to some point. Therefore, the person who has no stop loss has not really accepted any risk at all, because it's not really possible for him to lose.

[SIUYA 11]

  Predictor said:

This is a reasonable perspective. However, we are referring to a system in this case which had a historical drawdown and return. I may have should been clear that there are exit rules in a system like this. It takes losses: it just doesn't take stop losses.

 

Sorry...but this is clearly just a matter of accounting perspective - a stop or an exit form a system is the same. it is the point at which the entry trade is exited....how are they different? maybe its just a matter of definition but it irks me a little when things are discussed this way. For me a loss is a loss is a loss - no matter why, how or when.

 

  Predictor said:

I should probably add too and this maybe something new for those who haven't built system is that typically the tighter the stop one uses on a system then the GREATER the drawdown. Let me repeat, the tighter the stop then the systems will typically experience GREATER drawdown.

 

This I would say is completely system dependent (aka Zdo) - change the take profits to no take profits and test over different periods and you will get vastly different results. Plus when you talk about the greater the drawdown, you may have more stops per wins, but the drawdown may or may not change. I think you are being too general here. Often depending on the system your aim might be to build as many units as possible and run them, and stopping out at break even for example may lessen the number of trades on when you have a winner. You might then expect a big open profits drawdown far in excess of your closed PL drawdown.....for many this is a crucial element of the system, if it can be traded, if they can trade it.

 

 

  Predictor said:

My perspective in this article was that a person starts to trade a system that historically had DD of 40% or 50% because they know they will only have to risk x$ because the system uses a stop. However, they are really kidding themselves versus accepting the risk of taking that 40% or 50% loss.

.

 

yes...because records are meant to be broken the back tested drawdown is virtually guaranteed to be beaten at some stage, and the reality is they might stop the system at down 20%

 

(I dont mean to do anything more than post discusssions to the thread as I think too many go down the path of thinking a backtest is more than what it actually is and that often the same system can be radically different based on very small variations in that.....and this is I hope as you mention a thread for discussing the value of backtesting.)

[tommaso 10]

  Predictor said:

....typically the tighter the stop one uses on a system then the GREATER the drawdown. Let me repeat, the tighter the stop then the systems will typically experience GREATER drawdown. Of course, there is value in decreasing the maximum risk per trade because with less max risk per trade it is possible to leverage the system higher.

 

My perspective in this article was that a person starts to trade a system that historically had DD of 40% or 50% because they know they will only have to risk x$ because the system uses a stop. However, they are really kidding themselves versus accepting the risk of taking that 40% or 50% loss.

 

....QUOTE]

 

That is very true: a lot of people are deceived in thinking that stops limit losses, while instead may just be the opposite (depending on the general organization of the strat).

 

The only notable effect of stops in automated trading is generally to let your PNL possibly oscillate around 0 for a good amount of time before seeing larger losses ("slot machine effect").

 

[ Also consider (selling and buying appropriately) options, which may be a great way of hedging and margin bounding. ]

 

Tom