I believe in what Marco has done and the automated approach in general.
My concern NOW is that we are likely to begin to see this:
Note the word “placed” already. Or this:
Of course, Yuval is all over the map. It was not too long ago that he believed in Machine Learning so much that he was recommending Bootstrapping—even on spreadsheets if you had to (weeks).
Yuval recently told us that the way to avoid overfitting is by giving noise factors equal weight in a ranking system. This is just plain wrong.
Yuval has been known to aggressively argue opposing views within the same thread. Hard to know which way the wind will blow today.
But Marco is rock-solid and provides a rock-solid method.
You probably were not here when Marc was advocating adding errors and randomness to the data. He did not like us backtesting as much as we do. He reasoned that if he added enough errors to the data we would do something else.
I am here for what Marco does. I am not here to try to compete with the fundamental analysts on Wallstreet. And it is not possible to follow Yuval’s advice without him telling you that you are a fool for doing so at some point in the future.
If what Marco provides does not work it is because they are getting poor advice on how to use it, I believe. Marco will know when and how to update the Rank Performance Test. When it is time he should be given the resources to do so.
It is true that I have been an advocate for Marco updating sooner rather than later. I think I will shut up about that. I think I have been giving the Luddites (and people with math phobias) ammunition in their attempts to limit what Marco does.
From my limited experience with dynamic weighting, I have seen that most of the time there is a bias in some fashion. It could be as simple as mktcap or something much more complex. Weighting factors is what we rely on when forming ranking systems. So I think of DW as ranking a smaller universe. Ranking the best of the best if you will. I suppose it really comes down to trust. How much do you trust your ranking system, and how much do you trust your dynamic weighting formula? With the DW you can control how much trust you want to give since we can set our max and min levels. It’s another layer of tools that we have and I think it should at least be examined before being dismissed. It’s a shame to throw away a piece of gold because you thought it was a stupid rock while you were looking for diamonds.
There are several ways to approach this and it really depends how detailed you want to get. Where do the similarities end that you are comfortable with when doing a comparison?
Universe the sector or industry is the obvious one but that doesn’t always mean you’ll end up comparing apples to apples either.
Use a broad universe and look at the transaction history. Sort by sector or industry and see if there’s an outlier where your system consistently gets it wrong. Is there something unique about the outlier? Maybe the system is almost always wrong about REIT’s, hence why I’ve read that many exclude them. That could also mean there’s an opportunity for a REIT ranking system and you can include that in your buy rules if a REIT is selected.
You can make a bunch of small universes each with a ranking system and take the top 3 stocks from each into a book of 10 for a 30 stk port that are comparing more similarities than would otherwise be done on a broad universe. If you wanted you can still make a general ranking system and use it in each of the smaller port’s buy rules or maybe even in a DW.
Since size could be a comparison tool then of course you can use market cap as many do.
When looking for a TV you can fall down the research rabbit hole pretty fast since there are so many TV’s out there and they are all a little bit different. The question, where should my comparisons start and stop? It’s a very valid question but has a very broad answer. I just thought that it’s valid to consider it when forming a universe because the entire ranking process is based on comparisons. It’s easy to say I only want to compare apples to apples but an orange is still a fruit that is about the same size and calories. Should I include oranges? How about a banana? How about a car tire?
I don’t think you need to worry about liquidity if you were to publish your ranking system — just rebalance your system on Friday morning to avoid any liquidity issues on Monday morning.
This would take care of the liquidity issue, but it could create a legal problem since it looks a lot like “front running” subscribers. Whether or not an appropriate disclaimer (ie, stating the designer will be trading the method on Friday) would be sufficient to keep things legit with the SEC is something a lawyer could advise.
From my perspective as a sometimes subscriber to DM, I’d be OK with the DM creator trading the method on Friday (provided it was disclosed).
My assumption is that everything (fundamentals, estimates, and prices) gets updated each night (ie Monday, Tuesday, Wednesday, Thursday, Friday nights). We get to use this fresh daily data whenever we manually rebalance a portfolio during the week.
My understanding is that things are different when you run a simulation. The simulation has access to weekly data for fundamentals and estimates (ie the final value of the week) but does have access to daily price and volume data (so daily TA formulas can be used in sims). Only saving weekly data for fundamentals and estimates saves P123 a lot of storage space.
Am I mistaken? Could one of P123’s staff shed light on whether this understanding is correct?
o806, My understanding is estimate revisions only update weekly, but other price and fundamentals are daily. I’ve bookmarked the link below that shows update status.
Brian, even much better then a monthly update on! Once again on other platforms you will be able to do with higher controll (e.g. you can program
everything yourself), but first the data is the gold and how p123 is managing data (PIT, Price + Fundamentals) should not be underestimated, here is the gold!
Factset PIT → I had understood that it was PIT “good enough” - maybe not exactly the same PIT as S&P but likely to be decent and hard to figure out if one is better than the other. (I am aware that writing PIT “good enough” makes no sense from a theoretical / purist point of view. In theory there is only one version of the truth, but in practice …)
It’s a bit of a mess to me. Raw FactSet appears not to be PIT. Some consumers of FactSet data then make and keep data snapshots in order to construct a PIT database. I suspect the PIT moniker will fall into disuse here. You can’t sell what you don’t have.
o806, I might be wrong. Maybe that estimate update date only impacts the weekly figure that is used for backtests? Probably best to ignore me on this given your info.
Some of our estimates data changes weekly, some daily. For backtesting, we only use weekly data with the exception of prices. But for current screens and strategies, some estimates data will change from day to day.
Could you clarify what “some of our estimate data changes weekly, some daily” means? The reason I ask is I typically do not trade on Mondays but I would change my routine if Monday has the freshest estimate data.
Is the reason some estimate data changes weekly because there is no change in the estimate data on any day in the week — which would typically be the case if a stock is covered by only 1 or 2 analysts. In such a case, the estimate data would be “fresh” even though there no change from the previous weekend.
Or is the reason that your data provider has a systematic bias towards updating different types of stocks more frequently than others (ie large caps get updated daily but small caps get updated weekly because the latter has a lower priority).
Or is it that some types of estimates (eg for earnings) gets update daily but other types of estimates (eg for sales) gets updated weekly.
I’m sorry I was unclear. Your last guess is the correct one: some types of estimates data get updated daily but other types get updated weekly. The EPS revisions, for instance, are weekly only, while the EPS estimates are updated daily.
I hope I’m not giving the impression I’m trying to split hairs, but I’m still not sure if the Estimate factors I use are updated daily or weekly. But I really need to know since one of my systems is very heavily reliant on the freshness of the following factors:
CurFYEPSMean
CurFYEPS1WkAgo
CurFYEPS4WkAgo
CurFYEPS8WkAgo
CurFYEPS13WkAgo
and also
NextFYEPSMean
NextFYEPS1WkAgo
NextFYEPS4WkAgo
NextFYEPS8WkAgo
NextFYEPS13WkAgo
Are the above updated daily or just on the weekend?
Thanks in advance for your attention to this. I really appreciate it.
These are updated daily. But only weekly for backtests (like all our data except for prices and volume). The only estimates data that isn’t updated daily is the estimates revisions.