Hi Marc,
Thanks very much for taking the time to elaborate your thoughts on ETFs. They confirm what some of us needed to know, that ETFs are an incidental “because we could” add-on, not a core offering that will receive much development. My takeaways from your post…
Bottom line… There are not enough “EFT-only” investors to justify it. The ETF data delivers added usefulness to your core market, stock investors, and incidentally also provides some usefulness for those of us who are more interested in ETFs.
Understood. So there is no point in asking for anything that requires new data expense or considerable programming. I make the same decisions, too, in my business, and would make the same one that you have. And that’s OK…
Even “as is,” P123 does provides genuine value to those who are only interested in ETFs and who want to explore certain theories more deeply. ETFreplay seems to be the only other game in town, and while its interface is intuitive and its range of pre-set functionality is excellent, it does not enable one to customize beyond pre-set drop-downs.
Both have their particular usefulness. I wanted to push beyond what I was doing with ETFreplay, hence P123. I can also rough-quick-test some methodologies laid out in an academic paper faster in ETFreplay than in PI123, then decide whether to go deeper here. So they are both useful, each in their own ways.
For any who feel like me, we may be disappointed at not having more extended functionality. But a well-run company that does not chase the wrong markets means the continued availability of this excellent tool.
–
I don’t think, though, that your argument needed to go further beyond your business case. You lost me once you went there, first because we are now into a hypothetical that has been ruled out, and second because it reads as argumentation that is uncharacteristic of your writing, more to make a point than to present a fair and balanced argument regarding the potential value of the requests (assuming you could, as you said, “snap your fingers and make it happen”).
I agree with you that many ETFs, especially the newer ones, are merely marketing vehicles that add little (ex., “target date” ETFs for targeted demographics). And I agree that ETFs are top-down. But neither of those points obviate the benefits of having both high-level fundamental AND technical data upon which to screen.
Nor does the suggestion to use P123 to build and maintain our own rules-based ETF-equivalents (which is hardly as trivial a matter as you present it to be). As you said, we work top-down. Recreating 100s of ETF strategies from the bottom-up, just to test them, would be not only laborious, but artificial.
Some of your points selectively try to make cases that are actually more nuanced than presented. Yes, some asset correlations are converging. However, several macro-driven classes (ex, stocks, long-term treasuries, and gold) remain as non- and negatively-correlated as ever. The underlying drivers remain the same - prosperity vs contraction, high-interest vs. low, liquidity, crisis conditions, etc.
And I disagree with other points, such as the “even if we did this, there’s little benefit” argument. Your discussion of this shows that you look at this through a very different lens, missing many possible creative and barely explored (even in academia) possibilities, concepts that are potentially more powerful than what has been demonstrated anywhere, to date. The ability to back-test these new approaches, even if limited to the recent history of ETFs and hence not at the level of academia, would be most useful - I would invest in the most promising discoveries (needing only a sufficient cushion for a level of confidence).
I’ve shortened my reply and therefore oversimplified (hopefully not at the expense of over-bias), because there is no point in a detailed discussion about a rejected hypothetical. Your business bottom line remains the same…
Even if I am correct, there is nothing here that would not suddenly expand your market, leaving no reason to change your “no business case for this” answer, which I fully understand and respect.
I merely felt that some of your “post business case” arguments needed at least a bit of rebuttal.
The business case conclusion is what is it is, and it suffices. Every company has too many good ideas than they could/should ever implement, even Google and Apple. All we can do is allocate resources that best meet our core customers’ needs.
That, I believe, was your real bottom line.
And getting THAT right is hard enough. Rejecting goods idea does not mean that you or I need to feel that they were bad or useless ideas. There is simply more to gain by choosing to go with what your core market needs if the potential of a new market is not large. And on that note…
This thread is more powerful proof that your business decision is sound. Few people, only those interested in ETFs, spoke up here. Case closed.
–
Recognizing reality, please allow me to reduce the request to one suggestion which has no data cost and minimal programming time…
Benchmark of ETF Backtesting
The current benchmarks that are available cover pretty much every style, size and sector of equity, as well as all the basic indices. In a nutshell, stocks are well covered.
But ETF backtests cannot benchmark, say, a momentum strategy for gold against the price of gold (ex., the London PM Fix) – there is no such option in the Benchmark dropdown. Assuming you do not have access to that data, and since getting it would entail an additional cost, and since there are several other classes and sub-classes like this, here is what I suggest…
Please enable the user to benchmark against any ETF, IAU or GLD in the example above, by entering the symbol into a text entry box.
That would more than suffice. There are no new data costs, and only minor dev (since you will be using routines that are already programmed).
ETFs cover much more than stocks – other asset classes (ex. commodities, fixed income, etc.), specific countries and regions, methods, etc. Please let the user determine the single ETF proxy that best benchmarks the class, sub-class or even mix of classes that is being studied.
Benchmarking is a critical part of backtesting. Studies lose a great deal of their information and purpose without that ability. But we can’t do it for most types of ETF screens and their backtesting.
–
Thanks very much, Marc, for taking the time to express the views of P123 so clearly, and for considering the only request of significance that remains, one that costs very little, but would add much for “ETFers.”
With appreciation,
Ken
P.S. This comment was very interesting…
Being able to choose a proper benchmark would help. And yes, I do believe that we’ll come up with some original work. Scott and others are way ahead of me in that regard. There are some really creative and smart folks here.
But judging from the academic literature, 20-40 years of backtesting seems to be the minimal requirement in order to have a high degree of confidence in the results, which means publishing the statistical analysis to support one’s conclusions. But that raises a problem…
There are not enough ETFs with the longevity to cover full business cycles or to do tests that deliver higher levels of statistical significance.
So no matter what we develop with P123 for ETFs, there’ll be potential blind spots. That said, should we find an approach with a large enough “alpha cushion,” it’s worth putting into action at the practical level.
Card counting was being successfully used well before E.O. Thorpe, after all. It’s nice, though, to know you have the math backing you up.
So…
The development of recognized notable progress would require the presence of longer-term historical total return data, volume, and fundamental data on a range of indices that extend beyond U.S. stocks to such other asset classes as developed nation equity, long-term treasuries, credit risk bonds, real estate, t-bills, London PM gold fix, etc.
That would enable some truly ground-breaking work. I know that is even less of an option for P123. But it’s a direction that interests me…
Academics develop these systems using R, serious modeling software (Excel is also used). A programmer friend of my daughter, who used to program for a major financial firm, is looking to develop such a system for me in R, but it’s early days and I’m sure there are many pitfalls along this road. If anyone reading this has any interest or expertise in this, please PM me?
NOTE: Edits made to balance open and close tags of italics and bolds, fix minor typos. No change in content.