Declining Trend Trade Length Responsible for Recent CTA Losses?

[intradaybill]

@artemiso, excellent work! Man, you are fast!



Let me see if I fully understand your results: you have plotted the inverse of the difference from the mean trend length in bars for every ticker. Just to make sure, how do you measure the trend length?



Imo, a time-based graph would help to better see whether what Anthony alluded to is true. As I understand the claim is that during the last few years the trend lengths have decreased below the average. That could have also been true in some past periods. Another issue is that stocks like AAPL, AMZN, GOOG and others skew the results significantly with trend behavior that is not reflected in the indices.

[artemiso]

Thanks. I just have a half-decent architecture, the scripting is simple as braincell pointed out. (Edit)

[goldorak]

Did any of you wondered about the meaning of MA crossovers?



for example:



MA(5)=[ P(t-1)+P(t-2)+P(t-3)+P(t-4)+P(t-5) ] / 5

MA(2)= [P(t-1)+P(t-2)] / 2



MA(5)-MA(2)= [ 2*(P(t-3)+P(t-4)+P(t-5)) -3*(P(t-1)+P(t-2)) ] / 10



???wtf???



why not directly test ALL linear combinations directly?

[Baltazar]

Goldorak, I guess the intuition is more:

Ma are low pass filters, the longer ones, the lower the frequency cut off.

Difference between two low passed filtered signals give somewhat a band-passed filtered signal.

So by looking at the difference you get rid of low and high frequencies in the stock path.

Of course if you knew the interesting frequencies you could design from scratch a pass-band filter tailored to it.

[intradaybill]

@artemiso, thanks. Massive work you did!



I get your point about the summary statistics but imo a study that includes only the DJ stocks and their average trend length based on a 10/100 and a 20/200 SMA cross as a function of time may be more descriptive. In terms of trend following in the sense used traditionally I think the faster MAs are not relevant.