If you read the post where Physics Envy first arose you might notice that I strongly advocated for the validity of your ideas and for the value of studying finance. Without equivocation.
You suggested that I look up ‘Physics Envy’ which I did. I found one definition which I posted. From a fairly respected source that is certainly widely read.
You argue against the definition in the article I posted yourself:
You say you hope no one at P123 makes this argument but the authors at the NY Times do in fact make this argument. I disagree with them as you say you do.
I am sure there are other definitions we can both agree with but I did not see another clear definition in my short search. I suspect the term was originally used a joke but the authors decided to make a serious argument.
I don’t think I can be expected to edit the NY Times article by these authors to be a better article and then agree with it. The article as it is (and not how it should have been) I still disagree with.
I am not putting forth a hypothesis and substantiating it via backtest. In fact if you check the backtest for the Aerospace & Defense DM you won’t find a super-model over the last 20 years. That is the difference here. And if the next president is a democrat then I will suggest reducing exposure to the DM. It is not a case of “real-time testing”, but using my past experience in engineering to capitalize on what I believe is a political phenomenon. As for Cloud Computing, this is just common sense. If investors don’t want to invest in trends, that is their choice. I backtested these DMs, not to “validate” them but to provide an optimal strategy. Whether the strategy is optimal going forward remains to be seen, but there is no validation of hypothesis here.
I’m going to cut to the chase here. In your (or my) lifetime, how many legitimate hypotheses do you truly expect to have? 100, 10, 5? I would like to suggest that you will be lucky if you have one really good hypothesis in your lifetime. And that is if you are lucky. It might be along the lines of DCF for example, and even then success depends a great deal on how it is applied. i.e. most people will never succeed with DCF, yet the MOAT ETF is having some success because the people behind it apear to understand the limitations, plus they have hundreds of analysts running the numbers.
So if you have maybe one truly great hypothesis in your lifetime, who is to say it is “this one” in front of you today. I can come up with 100 different hypotheses (trust me, I am a clever guy), and maybe 10 will outperform in backtesting. Does that legitimize the 10 hypotheses? Because that is what we are really talking about here. Everyone is performing optimization and most of us don’t realize it. But if it makes you feel good then go for it. But don’t believe for a second that your process (hypothesis->test) legitimizes it. Because in all likelihood you are optimizing and probably optimizing to an extreme, unless you are the one person in a billion that dreams up a hypothesis and it tests out brilliantly.
Right, sure. This brings me back to Chris Columbus. He validated his hypothesis.
Jim - I wish I could get into your head!. Its gotta be pretty interesting there.
I know what I am going to say will offend a few people, but there is little in the way of fundamental factors and hypotheses that don’t find their origins in empirical data. The observations generally come first, then the explanations come to support the observations. Not much different than Einstein or Sir Isaac Neuton. The big advantage that Einstein had was that he didn’t come up through the University environment and therefore wasn’t stifled by prevailing theories. Let us keep in mind that his theory of relativity was never recognized by the establishment.
That reminded me of Gary Shilling. Since the '80s he has advocated being long the 30-year treasuries but always staying on the long end of the yield curve by rebalancing ever year. He claims to have done something like 5x better than the S&P500 over that time. That’s a once in a lifetime call and was based on the collapse of US interest rates. Google for more details since I haven’t checked his results in several years.
I think you give me too much credit. I’m not sitting here coming up with new ideas per say just trying to implement different facets of already established economic and financial theories. We obviously know backtests don’t predict the future with certainty nor do we expect a backtest to carry forward indefinitely or at all. With that said, I’m not sure I care to debate this any further because it is becoming quite pedantic. So I’ll leave it be from here. In the end all I care about is making money anyway.
Jim - as a healthcare professional, I’m sure you realize that pretty much every drug in existence is based on empirical testing and sometimes blind luck. (Penicillin comes to mind). How many substances are tested for carcinogenesis? How many substances are tested as a cure? How many researchers develop a hypothesis before they do any testing?
Correct me if I\m wrong, but only after an empirically-based observation is found does the researcher attempt to establish a causal relationship. Even then the researcher may not find or may not care to understand the exact mechanism. And how often is data (trials or analytical justification) biased by big business financials submitted to the FDA?
We are all surfing on the tide of financial information. We see the wave coming and stand up on the surfboard for a nice ride. Sometimes the sharks are out, sometimes we get thrown into the ocean, sometimes we get hit by a tsunami, but hopefully, the warning signal goes off in time. In the end, we hopefully get more out of the ride than it cost for the surfboard.
So for finance I am open to a lot of ideas including that the market is pretty efficient and maybe none of this really matters anyway. So let me just say I am open to what you might suggest and cede or stipulated to your ideas regarding finance.
With regard to medicine I will move to the FDA if you want (which would be the direct answer to your question). But let me move to what I know best, if I may.
Real medicine is practiced by interns (and not ophthalmologist). I was an intern for a year.
I started writing notes on my first day and on every patient I have seen, every time, since. The professor did not like my first note and was quick to correct me.
I am not sure exactly what I got wrong on that first note but he emphasized again that every patient, every time, has to have an assessment (multiple assessments almost always). I cannot count (considering ophthalmologist do this too) how many times I have written an assessment down. 20,000 times? Way too low I think. Closer to a quarter-of-a-million than 20,000. Edit. Okay a million times easy if you count the multiple assessments for each patient.
Of course, I am more certain of some of those assessments than others when I write them. Many times, I think, this assessment would easily meet any definition of hypothesis you might like to use.
But I am often ethically required to test this hypothesis (blood test, MRI, listen to their chest, put a finger up their rectum etc). And of course, I could get sued for “failure to diagnose” if I do not eventually get this “assessment” right.
That first professor let me slide for not doing the note right but go even a day without running or looking at a test of my hypotheses (assessments) and…….
Guess I might have survived switching into finance without testing hypotheses (after getting kicked out of medicine). Good to know.
I’m not sure what you mean by that statement but the research is done well before drugs get to the FDA as you are well aware. And research is often (or may I be so bold to say usually) done by testing numerous substances often at random to see if there is an effect on a particular disease or entity. And it is years of empirical testing before scientists get any kind of an understanding of what the property of the substance is that causes a positive outcome. That is if they ever understand it. Heck, that doesn’t stop them from treating patients even if they don’t understand the reason why it works.
But your assessment is based on experience and/or study of books, right? You have seen the sickness before. Somewhere in the past the sickness was part of an empirical study. If not then the doctor is lost. How often do you venture out into the left-field with a patient when something that you have not seen before or read about shows up? Do you create a hypothesis and test it? Or do you run a bunch of blood tests, do biopsies, etc. and hope to come up with a diagnosis? OK - if you are a brilliant physician then maybe you can generate a hypothesis and test for it. But chances are, at least on the TV series I watch, they will run a battery of tests and hope to come up with an answer.
Yes. That is because you live in the United States of Columbus. Those of us in Northern New India do not have nearly as many law suits, so not a big concern.
I don’t think finance works that way. It’s not a science.
The ratios and formulas we use to choose stocks are based in financial accounting practices. Those practices were developed over hundreds of years in order to make bookkeeping make sense. When people constructed the cash-flow statement in the 1980s so that the three main fields–cash from operations, investment, and financing–all add up to total cash and equivalents, was that based on empirical data? No. It was based on rules and customs. When you read a balance sheet or an income statement, all the numbers you find there are based on rules invented by some accountant twenty or three hundred years ago. What counts as net income and what counts as operating income and what counts as EBITDA and what goes into cost of goods sold and why R&D is not accounted for in the same way as CapEx–all that is based on convention. There’s nothing the least bit scientific about it.
Therefore all the ratios we use when we decide whether or not to invest in a company CAN’T have their origins in empirical data. Instead we use empirical data to test whether or not we want to use them. O’Shaughnessy tested ROE and net profit margin and decided they were practically useless, and when he tested price to cash flow and total accruals to total assets, he found they were very useful. But whoever came up with the formula for total accruals to total assets or the way we get cash flow by adding depreciation to net income did so based on accounting rules, not based on empirical data. Joel Stern, who came up with the idea of FREE cash flow (before him it was just “cash flow”), didn’t do so empirically. He understood how accounting works, and he understood that if you take net income and adjust it by adding or subtracting things like changes in net working capital and depreciation and maintenance capital expenditures, you’d get a measure of how much excess cash a company’s operating activities were generating. And he thought that that was a useful measure. Empirical data had nothing to do with it.
The difference between Einstein and whoever the geniuses of investing are is that Einstein was not observing complicated man-made conventions but instead was observing nature. The world of finance is nothing like the world of nature. It’s more like the world of high fashion, gourmet cooking, no-limit poker, or feng shui. It’s a very complicated system of very complicated rules that are all man-made and can’t be found in nature. Empirical data is very important to all of those worlds–god knows how awful gourmet cooking would be without empirical results!–but you have to start with man-made conventions. Observations come second.
You sure use a lot of bootstrapping for that view, IMHO. And you have recommended boostrapping to members. You used to use Omega which is probably best called a statistical tool and not machine learning. But a statistical tool used to help predict which stocks will do best for sure. You used to be the first link on a Google search for Omega.
What makes me think Marc might call it ‘Physics Envy’ if he were to notice me mentioning bootstrapping in the forum?
Not to mention I first heard from you that machines now play no-limit hold 'em better than the best pro. And you say no-limit hold ‘em is like finance above. So that could be expanded upon this before I get the points there.
You used to be a big user of regressions which is a statistical learning tool plain and simple. Not sure what you use now, maybe just bootstrapping and some correlations (a statistical measure).
You have clearly listened to our previous discussion in the forum when I mentioned the problems of heteroscedasticity. Or maybe you change your view on this later. Anyway you have recently said you do not like OLS so much anymore because of heteroscedasticity.
Surely quite a bit different than Marc’s ideas. Not to suggest any of that is Physic Envy.
Marc might not call backtesting and rank performance tests Physics Envy but you surely do it differently than he recommends. Or at least you have.
I do see bootstrapping as very similar to Jeff’s idea about adding random noise. Boostrapping is after all random selection (with replacement) and can be considered randomization as well as an ensemble (machine learning) method.
You use boostrapping for the same reason that Jeff was interested in randomization I believe.
Marc clearly did not like Jeff’s idea. He has not said anything about bootstrapping when you mention it despite the similarities.
Anyway not a feature request and perhaps appropriate to the conversation since you have an interest in the thread.
I get that you are in the same boat as we are. Doing a little with spreadsheets with some of the downloads and even being able to upload now thanks to P123. No different than all of us looking at some new ideas and doing the best you can with the tools we have access to.
Thank you for sharing what you are doing with machine learning on spreadsheets.
So not a feature request and not a complaint.
But I sincerely do not get the apparent conflicts.
This is where I stop reading. First, if it is not a science then it doesn’t make sense to apply scientific principals such as hypothesis testing. Second, the problem with the argument that accounting practices have been developed over hundreds of years is that it says nothing for the relationship between the company and the stock market. The ONLY semi-useful method of stock valuation is based on either discounted cash flows or dividends. And I say that begrudgingly because Marc brought that to my attention. And the problem with DCF and DDM is that the results vary dramatically based on the assumptions of the analyst. One is essentially trading one set of assumptions (such as P/E ratio or P/S ratio) for another set of assumptions that provide at least as much uncertainty.
Now what really gets me is when someone takes say 55 factors based on accounting principals and throws them all into a ranking system. But why did you choose those 55 specific factors? What makes you think that the blind combination of these factors somehow relates to accounting principals? I can understand if you took a couple of factors and rationalized how they work together. Maybe I could stretch that to 5 factors with a bit of imagination. But 55 factors, I’m sorry but you are well beyond accounting principals.Just for the record, I think that a large number of factors is the way to go. What I’m arguing against is the “ivory tower” arguments that I hear all too often here.
I told you I like you. Now I remember one of the many reasons.
“Ivory tower” bothers me a little too. But there is no question that one can find 'Physics Envy’ as Marc defines it going on elsewhere as you allude to.
I also like Marc. I was trying to think when he gave advice to me that was not heart-felt, appropriate and generally good advice. If he said something to me that I minded I have forgotten it.
If you want me to recount things that actually bother me and that I have a lasting memory about, I can tell you some more medical school stories and since you are interested…… Well, maybe you aren’t.
But should Marc ever decide to get upset about Physic Envy in the future there are others he can talk to as well as me.
I’m sorry that you stopped reading, because I don’t think we’re that far off from each other.
First, hypothesis testing is used in everyday life, not just in science. It’s part of basic problem solving, whether scientific or not. How am I going to make this omelette taste better? Well, I could add coconut flakes to the eggs, or I could add chopped prunes, or I could add parsley. My hypothesis is that parsley would work best. I’m going to test it. No science involved there.
Second, I’m sure that you use a lot more accounting stuff than the DDM or the DCF when evaluating a stock.
Third, I agree with you 100% that hypothesis testing is extremely important. As you rightly point out, a “blind combination of factors” is not an example of hypothesis testing. I spend about ten times more hours hypothesis testing than I do investigating accounting stuff. Maybe I should have a different balance between the two, but that’s the way I am. Statistics is important for hypothesis testing when you’re using tons of numbers. It can be easily applied to non-scientific stuff if one takes the right precautions. When urban planners model the effect of a new stoplight on traffic patterns, is that science? No. Does it involve statistical modeling? Yes. Statistical modeling is great for financial planning too.
In the end, I think where we differ comes down to two things. You maintain that finance is a science, I maintain it is not. You think that discoveries/innovations in the field of finance are data-driven and I think they have to be accounting-driven (because without the accounting there would be no data). But I think we agree on other points.
Wow. A rare thing for the forum. But I think I might agree with everything in your post, Yuval
I might want to say data is data and I am not sure I care personally where it came from or how it was derived. Of course, I want it to be data that somehow helps me to know what stocks to buy but I don’t care where it came form.
But I agree that we might use that data (wherever it came from) in almost the same way.
Maybe you say boostrapping while I say boosting. Different but not really.
If you could put in good word with Marc and suggest that at least with regard to the techniques we are both using a little ‘Physics Envy’ is okay it would be much appreciated.
Is your basic strategy to pick some growth trends, select stocks out of those growth trends, then create an optimized ranking system which backtested selections in those stocks in the near past? So somewhat of a hybridized approach?
I think that model making can include both accounting and data. Sorry I know that’s a little off topic of “physics envy”.
I do not agree that accounting drives data. If that where true then how could we explain the run that TSLA has had, as well as other extreme high growth stocks (Accounting could only take that so far)? Maybe data from momentum and price action would have foreseen that run-up better had we only looked in that direction with those filters? This is a part of the market and should not be ignored. However we could choose to ignore it and focus on the stocks that appear to be more accounting driven. I think that’s what most of us do because that’s what these tools at P123 seems to work best with.
With that said high growth stocks that move up or down based on very little accounting valuations are also possible to capitalize on and many investors love to play the long/short due to the volatility these types of stocks bring. Call it speculations, call it gambling, call it whatever you like but these stocks make up a corner of the overall market whether we like it or not. If we call it gambling then we can look towards betting methods of gamblers and that would be statistics … data.
There is room for both in the markets and if anyone agrees with that then we should accept that the market is not black and white and there are going to be stocks that fall somewhere in the middle. Maybe more stocks live in the grey zone than in either camp … I don’t know but I have to accept that it’s possible.
In the end all I think what matters is that we are able to find an area of the market that we feel has historically had correlations with either accounting, technical data or both. And do we feel there is justifiable merit that those correlations may persist into the future? Domain knowledge matters, fundamentals of corporate accounting, and valuation is just as important as technical data statistics. The more you know, the bigger your net can be when trying to catch winning stocks. I’ll never come close to knowing everything but that is no excuse for me not to learn more and try again.
I’m not a huge fan of putting labels on ideas. I think the world consists of a lot of grey zones and ideas get blended or even swapped at various points in time. Kind of like a market cycle going from a growth phase to a value phase.
I think I agree with you although everyone is making this more complex than it needs to be, I believe.
I mean isn’t accounting data, data?
English was not my best subject but ‘accounting’ is the adjective and ‘data’ the noun, I think. Hence it is data.
Let me try this. Important data. Yea, data I think. Important but still data.
Hard to get data. Still data I think.
I honestly do not get this. And sincerely, what am I missing?
Maybe the man-made part. My credit score. The whole concept is created by man. Still data that can be used in a lot of different ways. Maybe a linear regression of the amount in the bank and the credit score for consumers. This kind of thing is done all of the time.
Yep. Data and data that can be used.
SAT score. Pretty artificial and not from nature but people have sure figured out how likely you are to finish college based on your SAT score and whether to accept you into college.
Still not getting it.
Maybe that hypothesis thing: “If I get a perfect SAT score I will get into Harvard.”
Certainly a testable hypothesis.
Seriously not getting it.
But if Yuval is pretty much using this data like it is, well data then I will not get worried about it. Not getting it but not sure why I would care.
Jeff - that is essentially what I am doing. Ride the macro trend and hopefully catch the factors that are hot. Revisit every year as a minimum. If I don’t get the “right” factors then at least I am still riding the macro trend.
SteveA
Well, maybe you aren’t.