A technical indicator could give a buy signal for a security with one set of values, and at the same time, could give a sell signal for the same security with a different set of values! How do you trust the indicators then?! Moreover, if a set of values works this time, the same set of values may not work the next time!
Technical analysis uses historical movements of a security to predict a probabilistic future direction or price of the security. By definition, technical analysis is probabilistic and thus its predictions are correct sometimes and go wrong other times. And we are aware of this uncertainty and are perfectly fine with it!. However, it is not possible for an indicator to describe the accuracy of its prediction. In the absence of that, basing our trade calls blindly on such predictions is as good as basing them on coin toss results. This article examines ways to assign an accuracy number to the predictions made.
Given this problem statement, the first thing that comes to our mind is backtesting. The results of backtesting give an indication of how an indicator has fared in the past. It is important to note that backtesting shows different results if applied to different timeframes (between two dates) or with different sets of values based on market behavior of that period. Our goal then is to know which set of values of all possibilities best suits a technical indicator for a given security for the current market conditions.
We all know that the price of a security doesn't move in a straight line! It keeps moving in a wavy pattern making highs (crests) and lows (troughs), both of short and long forms. Not only does the price change, but also the frequency and period (distance between crests and troughs, or swing highs and lows) of the security change on a day to day basis! This dynamic period plays a crucial role in selecting values for indicator parameters. For, e.g., it would be inappropriate to choose a longer length moving average when the security is volatile and making shorter swings. Also, the right set of parameters keep changing for an indicator with ever changing period of the security.
Without going into the complexities of establishing relationships between period and indicator parameters, we could backtest indicators for all possible values to arrive at the best set to use. For simplicity, let us consider a strategy that gives a buy signal if slope of the simple moving average is positive and sell signal if the slope turns negative. All that this takes is a single parameter - length of bars to the past (moving average length). Let us backtest with different lengths of and plot P&L for each with probabilities based on the number of trades won. The length with the best P&L could be considered as the ideal parameter. Note from the chart how this changes over time. Note also that this changes based on backtesting lengths, the chart uses 3-Jan-22 as the start date. (Another Note! Approximate P&L calculations with both long and short, for demonstration purposes only)
In summary, technical analysis methods work well with the right set of parameter values. And choosing the right set still has a lot of uncertainties to it, though the uncertainty could be reduced by backtesting and choosing a better set from time to time.
Technical analysis uses historical movements of a security to predict a probabilistic future direction or price of the security. By definition, technical analysis is probabilistic and thus its predictions are correct sometimes and go wrong other times. And we are aware of this uncertainty and are perfectly fine with it!. However, it is not possible for an indicator to describe the accuracy of its prediction. In the absence of that, basing our trade calls blindly on such predictions is as good as basing them on coin toss results. This article examines ways to assign an accuracy number to the predictions made.
Given this problem statement, the first thing that comes to our mind is backtesting. The results of backtesting give an indication of how an indicator has fared in the past. It is important to note that backtesting shows different results if applied to different timeframes (between two dates) or with different sets of values based on market behavior of that period. Our goal then is to know which set of values of all possibilities best suits a technical indicator for a given security for the current market conditions.
We all know that the price of a security doesn't move in a straight line! It keeps moving in a wavy pattern making highs (crests) and lows (troughs), both of short and long forms. Not only does the price change, but also the frequency and period (distance between crests and troughs, or swing highs and lows) of the security change on a day to day basis! This dynamic period plays a crucial role in selecting values for indicator parameters. For, e.g., it would be inappropriate to choose a longer length moving average when the security is volatile and making shorter swings. Also, the right set of parameters keep changing for an indicator with ever changing period of the security.
Without going into the complexities of establishing relationships between period and indicator parameters, we could backtest indicators for all possible values to arrive at the best set to use. For simplicity, let us consider a strategy that gives a buy signal if slope of the simple moving average is positive and sell signal if the slope turns negative. All that this takes is a single parameter - length of bars to the past (moving average length). Let us backtest with different lengths of and plot P&L for each with probabilities based on the number of trades won. The length with the best P&L could be considered as the ideal parameter. Note from the chart how this changes over time. Note also that this changes based on backtesting lengths, the chart uses 3-Jan-22 as the start date. (Another Note! Approximate P&L calculations with both long and short, for demonstration purposes only)
In summary, technical analysis methods work well with the right set of parameter values. And choosing the right set still has a lot of uncertainties to it, though the uncertainty could be reduced by backtesting and choosing a better set from time to time.