8 Comments

Thank you, Luke! I think you could make your job a little easier in the future via QuantConnect, which lets you algorithmically test various strategies starting from the 1980s.

Expand full comment

Oh cool - I hadn't heard of that before. Thanks for the heads up : )

Expand full comment

Joe brought up good concerns. Another concern with this methodology is that it measures (cherry picks) across a long (mostly) bull run period. This further skews the sample.

Expand full comment

Agreed!

One thing I wonder is whether that bull-market regime that has held since 2009 is truly over. We are seeing a lot of strange things in the market that might suggest that to be the case. If that bull run did end on January 3, 2022 and we are in a new downturn, how far will the current market drop? How soon will it be until the next bull run begins? And, how long will that one last? All important questions to consider.

Expand full comment

It is difficult to get an accurate view of this strategy if only using a list of current S&P 500 stocks. Ideally, the data needs to include all stocks that have moved in and out of the index and it needs to distinguish when those changes are made. Only then can you have confidence that the list of S&P 500 stocks is truly point-in-time accurate. Otherwise you get a lot of survivorship-bias. Also need data that includes dividends and capital adjustments.

Expand full comment

Hi Joe - I completely agree. I hope my caveat at the end was strong enough to convey that this is a pretty simple experiment.

Additionally, I wouldn’t have posted this if I didn’t perform the same analysis with only the stocks that have remained in the index going back through the full horizon. That’s not a perfect approach but does help a little bit and the results were largely the same. The level of outperformance is what stood out to me - and it’s not just driven by one stock but fairly noticeable across the ten.

I also would need to go back further than just the last ten years but that is difficult to do without the proper dataset as you have mentioned since the survivorship bias would only become more exacerbated with a longer time horizon.

Thanks for the note!

-Luke

Expand full comment

Yes I guessed that you knew this but I wasn't quite sure what you meant. I've done tests like this in the past (using paid data sources) and buying winners does generally seem to work, certainly better than buying losers.

Expand full comment

It feels like it should be the opposite though - at least in my gut haha

When I think about the whole concept of overvalued and undervalued I’d tend to say that winners and losers should have some reversion to the mean. So, when I see the opposite it feels like something is wrong to me. But, I guess this is why a momentum factor is a thing!

Expand full comment