Man this research is definitely taking longer than expected, but I am making good progress so I guess I can spare my spare time for this. Although I would like to finish it, every time I find more interesting stuff out there.
In this article I want to talk about the Exponential Moving average, and how to use it properly, and what features I have found out about it.
I have talked in the past about the EMA, but mostly negatively, although further research proves that itās not that bad, actually if you donāt have any other tool in your trading software or such, then definitely sticking to the EMA is not bad.
However people were using the EMA incorrectly, let me show you how to use it properly.
The EMA
So the EMA as any other moving average, is trying to estimate the mean of the probability distribution of the price. Since itās heteroskedastic, the distribution will have a āmoving meanā or a ātransitioning meanā rather.
This means that the mean is not the same. With a coin toss, itās always 0.5 since itās a static event. The market is not static, itās a moving time series, so the mean will transition as well.
If the market were efficient, then forecasting it would be impossible. But itās not. There is always a little bit of inefficiency there, just enough to exploit it with the proper tools.
If the market were efficient, then the best forecast for the next datapoint would be the last datapoint.
However since itās not efficient, thus incorporating several previous values into the forecast equation is essential, it gives us more information, since the past datapoints carry potential future information with them.
So the mean will definitely not be an average with equal weights, like the arithmetic average, we will have to assign weights to each LAG according to their importance in the equation.
Obviously the latest lag will bear the most weight, and slowly or quickly decay afterwards into insignificance as we scoop in the past for information.
How to assign the Weights?
This is the only question that matters. Now I wonāt reveal here the exact methods how to use them, obviously the model can be more complex. But just as a basic overview to market newbies to guide them towards better accuracy when forecasting the markets.
Most of you have been using the EMA weights totally wrong. In most trading softwares the EMA is defined like this:
So the variable is traditionally called āalphaā and it ranges between 0 and 1, where zero means no EMA and 1 means no weight on the previous values, essentially a random walk model.
If we set the the alpha to different values we can see how the weights and the EMA changes:
Alpha = 0.1
Alpha = 0.5
Alpha = 0.9
As you can see the higher the alpha the more emphasis on the previous value, the less emphasis on the ones earlier, decaying quite quickly.
If the Alpha = 1 then only the previous value is taken. This signals a random walk, as this information is useless from a forecasting standpoint, we donāt learn anything from this model.
Which Alpha is optimal?
Well if the model sucks, then the value of 1 will always give the biggest results. In the snippet above, the 0.5 value seem to give the lowest error. However on larger samples and simulations, the correct value will always converge towards between 0.9 and 1.
On a trading software this is impossible to set. Since the trading software will follow the 2/(PERIOD+1) formula, even if you choose a PERIOD 2 based EMA that will be an alpha of 0.66.
And of course a PERIOD 1 will be 1, and that is not good.
You literally canāt choose a 0.9 for Alpha on most trading softwares, yet all evidence points towards the optimal values ranging in those ranges.
Itās almost like a conspiracy, that the brokers and software makers who design those tools donāt want you to have accurate tools. Interesting isnāt it?
No wonder they put all kinds of crappy useless Technical Analysis tools that donāt really work.
How to use the EMA?
Again, you were probably been using it wrongly. Itās not meant to be used as crossovers, itās just a forecasting tool. The Nth EMA is the N+1 Price, if itās configured well, then it should give an accurate forecast for the N+1 Price datapoint.
Again, itās not just as simple as a basic EMA, there can be other elements added to it, like a volatility estimator like a GARCH or an error correcting term.
More sophisticated quantitative tools give better results, since it factors in other properties of the price.
But with just a basic EMA alone set between 0.9 and 1.0 the forecasting accuracy will increase massively.
The Alpha of course has to be estimated at every new datapoint, or using an adaptive estimator to make it āautopilotā.
Conclusion
So it looks like I will be using the EMA as well, or at least a variant of it. It is definitely useful as a forecasting tool, but itās just a basic one. There are much more advanced tools out there, with much better accuracy.
Disclaimer: The information provided on this page or blog post might be incorrect, inaccurate or incomplete. I am not responsible if you lose money or other valuables using the information on this page or blog post! This page or blog post is not an investment advice, just my opinion and analysis for educational or entertainment purposes.
Sources:
https://pixabay.com
https://www.pexels.com
