Average tester – are averages any good?
- This topic has 15 replies, 4 voices, and was last updated 3 years ago by
.
-
I have always hated averages. They are horrible lagging things that can easily be curve fitted but I decided to code something to see a little more easily how bad they are and to see if any of the popular average types are better than others.
So I coded this simple code that uses arrays to record the average gain/loss of any candle if it is preceded by a candle that meets a preset average condition. This condition can be something like close > average or low > average or even fast average > slow average. This tests long conditions but you can also test short by using the reverse condition.
The 8 most popular average types are tested so that their performance can be compared. The result is the average gain per candle. A datum of every candle on the chart is also calculated so we can see how our average condition performance compares to buy and hold.
1234567891011121314151617181920212223242526272829303132333435363738394041424344454647//Period setting for average//p = 24if barindex > 1 then//Choose only 1 of these result conditions//c1 = close - open //for long//c1 = open - close //for shortif islastbarupdate then$lastopen[0]=openendifif open<>$lastopen[0] thenfor t = 0 to 8//Choose only 1 of these entry conditions////Long Conditions//avgcond = low[1] > average[p,t](customclose)[1] //Low is above average//avgcond = close[1] > average[p,t](customclose)[1] //Close is above average//avgcond = average[p,t](close[1]) > average[p,t](close[2]) //Average is risingavgcond = average[p,t](close[1]) > average[p*2,t](close[1]) //Fast average is above slow average//Short Conditions//avgcond = high[1] < average[p,t](customclose[1]) //High is below average//avgcond = close[1] < average[p,t](customclose[1]) //Close is below average//avgcond = average[p,t](close[1]) < average[p,t](close[2]) //Average is falling//avgcond = average[p,t](close[1]) < average[p*2,t](close[1]) //Fast average is below slow averageif avgcond then$count[t] = $count[t] + 1if c1 then$total[t] = $total[t] + c1$avg[t] = ($total[t]/$count[t])endifendifnextendifdcount = dcount + 1dtotal = dtotal + c1davg = (dtotal/dcount)endifreturn $avg[0] as "Simple", $avg[1] as "Exponential", $avg[2] as "Weighted", $avg[3] as "Wilder", $avg[4] as "Triangular", $avg[5] as "End Point", $avg[6] as "Time Series", $avg[7] as "Hull", $avg[8] as "Zero Lag", davg as "Datum"The first attached image is of a test on the DJI daily chart. It tests long performance if the 24 day average > 48 day average. All averages have a lower average gain per candle than the buy and hold datum except for Hull, Zero Lag, Time Series and End Point averages where the performance is improved. So perhaps a simple filter like this can improve a strategy?
The second image tests if close > 24 day average and only the Time Series average beats buy and hold.
1 user thanked author for this post.
Well facts are facts. It is possible to add an average filter to a strategy and that filter mean that we are long on a market or short on a market during periods that statistically would have in the past given us an advantage (i.e more upward or downward pips per candle than the norm).
What I think is also interesting in the small amount of testing that I have done is how useless comparing price to an average is compared to using a fast average and a slow average.
Also pretty much all the time simple moving average, exponential and weighted averages are pretty useless as a filter whereas Hull and zero lag seem to fair much better.
No average filter will stop you from trading in a market when that market is going to lose you money.
So until I test further I would say that there is a place for averages in trading but I think it is limited.
1 user thanked author for this post.
Find exclusive trading pro-tools on 
