If you have traded for long enough, you have probably found yourself discussing at some point the issue between price action and technical indicators. Some traders – after a long battle with indicators – find a sense of elegance and simplicity in price action analysis while others argue that indicators provide for information that cannot be obtained from simple price action. Is there a fundamentally better way to understand the market? Is there any advantage you can get from using price action? Is using indicators a way to ensure that you only produce systems that work in hindsight? On today’s post I seek to answer these questions as well as give you some clear facts regarding both price action and technical indicators. We’ll go through the definitions for both, how they are different, how they are the same and how these characteristics might relate with actual trading and profitability.
–
–
Let us start by defining both concepts. Price action is generally defined as any interpretation of the market that arises from simple comparisons of raw market data. For example if your trading system compares today’s open with the close 10 days ago then you can say that you are currently trading a price action based system. Price action strategies do not process raw market data in any way and therefore provide the purest image of the market. Technical indicators, on the other hand, process raw market data in some way that may eliminate some information contained within the original data. For example a stochastic oscillator gives you the location of current price relative to the High/Low range for the past X periods and in the process it reduces all market information to a 0-100 range. Technical indicators go through data and return values that are an overall simplification of the underlying price action. You could view price action as a loss-less way of trading while technical indicators are a filtered way of trading.
From the above it could be easy to say that price action is better because it provides more information. However, the amount of information in the market that is relevant for predictions is small, as the market is incredibly noisy. For this reason technical indicators provide a level of filtering that can be very useful in deriving systems that generate signals based on the actual underlying market behavior that is interesting to us, rather than the noise that is above. Indicators become more useful as the amount of noise in price action becomes larger (as time frames go lower) while they become less useful as the amount of noise becomes less. While a price action based system may have a very hard time producing viable signals in a lower time frame, a strategy based on technical indicators may be able to derive better results by simply looking beyond a lot of the noise that makes price action trading harder.
The above also does not mean that technical indicators become irrelevant as the time frame becomes larger. A technical indicator can provide you with information that is difficult to see across large amounts of data. For example a technical indicator might be able to show you what percentage of the past X bars where bullish or bearish, something that is difficult to deduce from simple price-action based comparisons. This also does not mean that price action is irrelevant at low time frames as price action can often react quickly to some events that are difficult to see through indicator filtering. While an RSI filter might take some time to react to a market spike, this can be caught very rapidly with a price action based strategy that is making some quick OHLC based comparisons using data from only the past few bars.
From an algorithmic system creation perspective people often complain against technical indicators because they are perceived as being more prone to “curve-fitting”. People often think that their chances of developing a successful strategy with indicators are slim because the degrees of freedom inherent to the indicators themselves – the variables relevant to their signal processing – gives them the ability to adjust the system (curve fit it) to past market conditions in what is perceived to be “excessive”. However price action based systems can do exactly the same thing if they are provided with enough degrees of freedom. A trading strategy based on 10 price action based rules can actually be more prone to being “curve fitted” that a strategy that is developed using a single indicator with 3 filtering variables, simply because the price action based strategy has more degrees of freedom. The issue here is related to the data-mining bias of each one of the two approaches, provided you have determined your data-mining bias for whichever number of degrees of freedom you have there is no inherent advantage or disadvantage inherent to the type of system you’re using. The issue here is related to the degrees of freedom of your strategy and not to the type of variables being used.
–
–
It is also worth mentioning that I have had experience in the past with both types of trading strategies (based entirely on price action and entirely on indicators) and I haven’t found any indication that one source might be better than the other. A strategy can be developed in an entirely sound way for any of these two approaches and neither one nor the other gives the strategy an additional probability to succeed under live trading conditions. In the end both price action and indicators are sources available for the design of trading strategies and a smart trader would definitely take advantage of both to develop better trading systems. Indicators can be used as signal filters to get information that would otherwise be difficult to get, while price action can be used to provide additional confirmations, fast reacting exits, etc. Both of these sources have their uses and they can indeed be combined to arrive at a more holistic approach to trading.
Finally I would like to point out that it is very important to understand indicators whenever you use them. Indicators are signal filters and as such you should understand what they are filtering, what information they are giving you and how you can use this information to create or improve a trading system. Indicators can rarely be used successfully without a deep understanding about what they are calculating, how they are being calculated and what these calculations say about the underlying price action.
If you would like to learn more about system development and how you too can develop your own portfolio of systems for algorithmic trading please consider joining Asirikuy.com, a website filled with educational videos, trading systems, development and a sound, honest and transparent approach towards automated trading in general . I hope you enjoyed this article ! :o)
Hi;)
Very nice article once again and I completely agree, price action alone is nothing better than an indicator based system. Indicators also use the price action as their inputs and it´s for example essentially the same if you look at the last 5 minutes of price action with your eyes and make a descicion based on that use let a 5 period long moving average on the 1M chart make the descicion for you. Both use the same underlying price action, both can be valid or not, however you interpret them. There is no advantage for one over the other.
Daniel, you say “.. simply because the price action based strategy has more degrees of freedom. ”
How do you define “degrees of freedom”? For the RSI(n) > X you have infinite possibilities because X is a real number. Price action has always finite possibilities. You fail to see that indicators on real numbers produce infinite possibilities. Even in the case of a moving average cross of the type ema(a1) > ema(a2). The possibilities are infinite here because abs(ema(a1) – ema(a2)) is a real number. All price action derivatives have infinite degrees of freedom and therefore pose the danger of uncontrollable data mining bias. Contrary to that price action degrees of freedom may be large but are always a finite number. Think about it…
Hi Bob,
Thanks for your comment :o) It’s an interesting point, but this involves only cases where indicators are evaluated against a threshold or contain a degree of freedom that is a real value. For example you could have RSI(a1) > RSI (a2) and you would have no comparisons against a real number or any infinite degree of freedom involved in the calculation of the indicator. I also don’t think that the RSI(n) <> X case is unsolvable as well because you can determine the data-mining bias by using the grid size you use to mine X within the system creation process. Clearly if you’re mining an RSI system with a 0.00001 step in X you need to use this value to determine your data-mining bias as well, which will obviously be much larger. The data-mining bias can be determined by the data-mining process you are going to carry out. Clearly you will be mining X within some step boundaries and within some degree of precision.
Your observation does imply that one needs to be careful when using indicators for data mining but I believe it does not rule out the possibility. In any case, always happy to read your comments Bob. Thanks again for commenting :o)
Best Regards,
Daniel
” For example you could have RSI(a1) > RSI (a2) and you would have no comparisons against a real number or any infinite degree of freedom involved in the calculation of the indicator. ”
The point is that the set of indicators involves infinite degrees of freedom although specific examples may not. RSI(a1) > RSI (a2) is rarely used but RSI(a1) > X is frequently used and this has infinite degrees of freedom although you can define a smaller step to try to mitigate that. The point is that in principle there are potentially infinite degrees of freedom involved with the use of indicators.
Let us consider an example. A trades 3 bar price action and B trades moving averages and he needs at least 25 days for a 5/25 cross. A has 6 degrees of freedom. B has an undefined number of degrees of freedom because we do not know how he came up with the 5/25 cross. Let us assume that he tested from 15 to 30 for the slow and 3 to 10 for the fast. There are 35 degrees of freedom in the best case.
I think your analysis was done in haste. These things require math and examples. I expect more from you than unbacked arguments. You have to live up to your legacy.
Hi Bob,
Thanks for posting :o) Let me now comment on some of the things you have said:
I think you’re confusing degrees of freedom with possible permutations. A moving average strategy with a fast and a slow MA has only 2 degrees of freedom (the periods) independently of the number of steps or ranges that are used within the optimization process. I am using the definition of degrees of freedom as formally given in statistics (the number of degrees of freedom is the number of values in the final calculation of a statistic that are free to vary, see here). In the case of a price action strategy using 1 rule there are many more degrees of freedom if you consider the shifts of the candles and whether the High, Low, Open or Close are used (4×2, for a total of 8).
In this sense your previous RSI example has only two degrees of freedom (the RSI period and X) although the number of possible permutations for X might be infinite. A degree of freedom is simply what can vary, not in what measure that variation can take place. However considering the number of permutations is fundamental for the determination of the data-mining bias (as I believe was your point?). If the permutations are potentially infinite you’re right in that there is ambiguity in the data-mining bias because we can never be completely sure of what the number of tested permutations were. A person using an RSI > X strategy could have tested 1 million or 200 combinations, we can never know for sure because the setup allows for that.
Thanks for the tough love Bob ;o) I’ll keep this in mind for future posts,
Best Regards,
Daniel
“think you’re confusing degrees of freedom with possible permutations.”
Not at all. In a simple optimization of a single hypothesis you are correct that a crossover has 2 degrees of freedom. In the context of data-mining, which was the original context of your post, each permutation is a degree of freedom. This is the case of a permutation playing the role of a variable in data-mining. Each of the permutations resulting from changing those variables becomes in turn a degree of freedom of the data mining process.
Is this so hard to understand? :)
Hi Bob,
Not at all, I agree with you in that the data-mining bias can essentially be infinite for strategies involving real numbers, as you cannot know what constraints were used in their creation. However I do believe that if you’re the developer you can come up with indicator strategies with limited data-mining bias that can work just as well as price action ones. I still don’t think that indicator strategies are inherently worse, provided you control the degrees of freedom and the data-mining process (you control the whole development). I see no reason why you cannot come up with sound indicator based strategies that are equally controlled as price action based ones (for example if you treat indicator series as price action series and perform only X(a) > X(b) comparisons). Sure, you need to be more careful and your data-mining space could be much larger but you can still do the process soundly provided you account for all the above.
Indicators are certainly prone to more pitfalls, but this doesn’t mean that they cannot be used in successful data mining approaches provided all problems are properly assessed to control the data-mining bias.
Thanks again for participating in the discussion Bob :o)
Best Regards,
Daniel