Hi – Yes, I don’t have most of them live not because none of them showed promise but for various other reasons such as the need for more demo testing (sometimes I forget to extend the demo validity – annoying that it’s only 30days) , more data analysis and simply in some cases, not enough time as more focus is taken up on shorter tf strategies. But I have it on list to revisit. In the meantime, I came across these short observational statements in my reading today that I think sums up nicely the purpose of this thread and echoes exactly my feelings and delivers it more succintly :
“More data intuitively feels better but things change over time, so more data can add in past trends that tell you nothing about what to expect in the future. If something is rare, the tendency is to go way back in history to gather as many examples as you can. But the further back you go, the more likely you are to study a world that operated in a different paradigm. We can study the financial crisis of 1873, but the modern banking system lives in a totally different universe than it did then.”
“Ask an economist how many recessions they wish they had to study and they’ll say “ten thousand.” But there have only been 11 recessions since we started systematically collecting good data in the 1940s. That is nothing. If we had as much recession data as we wanted we’d likely realize that half of what we think we know about recessions is wrong.”
“The most important events are the hardest to sample, so extrapolating their lessons can be misleading. There have been two world wars in the last century. But no one would say, “the average world war lasts five years,” or “the average world war kills 60 million people.” No forecaster would say we’re overdue for a world war based on historic trends. A sample size of two can’t tell us what to expect next. The point may be obvious with world wars. But use a slightly more common case and people start drawing all kinds of false conclusions. Studying things with inadequate samples causes more delusion than intentional deceit. That’s because the ease of explaining things based on what’s readily in front of you meets the intellectual cover of using statistics.”
I personally think a large sample size of data (>15yrs) is perhaps most important when looking at seasonality driven markets as they add some definitive value to the analysis, the most obvious example being commodities. But for most other markets I would question the use of a larger dataset as it will inevitably lead to false positives that will likely not work in todays markets.
.
