I’ve been happily trading some R2G’s but I’m wondering if I’m leaving money on the table. Are there advantages to private ports over R2G’s? I’m wondering if should I upgrade my subscription and build private ports. I’m asking the pros who have been doing this a long time and run lots of private ports.
My thoughts of advantages of private ports:
-Might generate more alpha with fewer than 5 stocks
-Allows me to pick more illiquid stocks
-Traded by fewer people
-Can private ports rebalance daily? (This might be the determining factor for me)
-Any other advantages?
I know that fewer stocks and more illiquid stocks can be more risky. I’m looking for high-risk high-reward for a small portion of my investable net worth.
The only real downside is that lots of time is require to build and test these models. The models I built during the trial period were good but no where near as good as many of the R2G’s.
I would not underestimate the amount of time that it can take to build models as good as some of the better r2g. The test for these models will be if they can perform as well out of sample. Tom previously mentioned that he spends 20 hours per week on this site and there may be others that spend more. Your learning curve will be based on your financial knowledge, programming skill, and ability to avoid over-optimizing (the latter you probably have based on your engineering background)
I would never understimate an engineer or anyone with a strong background in math as some of the best
investors have a similar background, like Renaissance Technology (my background is applied math)
I’ve been doing this for 8 years, and to really beat the market is much harder than people realise.
We all want a trading system that does better than picking stocks by throwing darts at a dartboard (random selection). However, here is the problem:
Dartboard portfolios may do well, may do very badly, purely by chance. The lower the number of positions, the wider the range of possible outcomes. We obviously don’t want to rely on luck otherwise we would be down at the racetrack!
If we have a computer that can run simulations of a million dartboard portfolios, then the best performing one will look amazing, astonishing alpha, call this our “prime” model.
We then analyse our prime model and conduct some “robustness” tests, and discover the probability of the portfolio performance being attributable to pure luck is one in a million. We conclude we are really onto something, put it into action, and see it fall apart.
The problem is that although we correctly calculated that the model was a million-to-one shot, we failed to consider the fact it had come from a million tests, and the result really could have been just a lucky outcome (as opposed to finding a genuine stock picking algorithm).
There are lots of random effects that push stock prices, mergers, CEOs, economy, contracts, fraud - things that cannot be accounted for by the quantitative approach. Distinguishing between the lucky model and the working model is hard, unless you limit the degrees of freedom (i.e. don’t run a million permutations) and keep the number of transactions high.
This may be an advantage as they fly more under the radar. As a small investor you can cease opportunities the large guys know about, but cannot extract. Beware they are volatile so you need more positions for a robust portfolio.
There are some serious issues with daily rebalancing. The ranking is done weekly in simulation, but can be done daily in a live port. Pricing data is available daily. If you want to have daily re-ranking, this cannot be simulated.
I’m finding that once I find a good ranking system and see that it works out of sample, I’m looking for ways to expand the use of the ranking system: more stocks not less. E.g., expand to 10 stocks and drop that other lame port (what was I thinking), different universe: large caps? Canadian stocks? Use InList? Drop the momentum factors and run a different port: are the stock the same or does the new ranking system pick different stocks?
I do not want to go from 5 to 4 with a good ranking system. I want to go from 5 to 15.
Going lower than 5 stocks will increase beta and volatility much faster than it will increase alpha. You may even be getting above the optimal Kelly Criteria bet size if you don’t have other ports: very bad.
@pdemartino. Thanks. Can you clarify whether you use close(0) to construct value ratios. E.g., Price to Sales (Pr2SalesTTM), Price to Free Cash Flow (Pr2FrCashFlTTM) etc?
How much is your time worth? Time at work? Relaxing? With family?
Building and backtesting systems take a LOT of time. If you’re working full-time…not sure it’s worth it. The major benefits are:
Emotional confidence to stick with the system over a long-period of time. Including losing years.
For me, this is the single biggest reason I trade my own systems. I know what I’ve done. What’s under the hood. I know the range of returns I’m expecting (which isn’t the prettiest backtest sim. I’ve ever built on it). And I know the role of the system in my overall portfolio.
I would not trade fewer than 5 stocks in any system. I would not trade fewer than 20 stocks in any ‘return driver.’ Return drivers are hard to find. A min. of 20 stocks (ideally somewhat more - say 30) is what I feel I need to reliably track a return driver over time. (So…if I traded a 5 stock system, and I rarely do…I would need 4 variations of it to give me any real shot at capturing that return driver).
I would not trade more than 10% (5% target) in any system.
Is it fun for you. Do you love it? It’s not a bad hobby. But, if you’re treating it as a hobby, your asset allocation weight to your systems should be in line with that.
I don’t do any daily trading. Once a week is plenty for me. I’ve reduced my trading over the years.
Building and trading can be a HUGE distraction from other things in life. Have to commit a lot of money and time to trading to make the ‘return’ worthwhile.
You could fairly quickly build a basket of systems from public sim’s and information. Could be divided as follows:
Piotroski and other fundamentally focused ‘balanced systems.’ If you have small port, or only want a few systems, probably start here.
Value based systems. Maybe with a ‘small tilt’, maybe market cap agonistic.
Technically focused systems. The most speculative. The highest potential upside.
Simply trading equal weights with a value tilt will allow you to beat many pro ETF’s. Adding a filter or two to eliminate ‘bad companies’ can help improve that.
But…depending on your portfolio size, you’d probably do better if you enjoy the rest of your life and work, to look for a bunch of systems from ‘good designers’ (however you define that) and/or public sim’s and put in 10-30% of your money across a basket of them, while leaving the rest in a balanced portfolio of ETF’s.
@Tom. It really is work. Time versus reward. I was running some statistics on one of my systems the other day. I noticed that I was making about 50 dollars per round-trip-trade ($48 dollars on one of my better high turnover systems). Or $24 dollars (on average) each time I enter a buy or sell order. I guess I can do a few of these in an hour and there is the hope of that compounding thing but not a magic formula.
Thanks for the feedback guys. I am willing to put in time, but not if the alpha, on top of good R2G’s, is small.
For now I’m just assuming that my strategies are sound and I avoid all the pitfalls such as data mining, overfitting, etc. I’m looking at more what P123 can offer and if it is possible to extract more information and thus more alpha. For private vs R2G.
Say earnings comes out on Wednesday. Does P123 know about it right away? Does P123 know it on Thursday morning? Or does it have to wait until Sunday night / Monday morning?
Another thought is that some of the stocks getting picked by R2G’s fluctuate a lot from Monday morning to Monday morning. It’s a whole week of lots of price action that is being ignored. It’s plausible that a value stock on Monday has moved up so much that by Tuesday it has moved down the ranking system and is no longer such a great value. Do you guys see this at all? Why not sell this stock on Tuesday and replace it with a better one? Not even necessarily generate higher turnover, simply allow more flexibility by re-ranking daily instead of weekly. This should be worth doing, right? Same basic strategy, just high frequency.
Also, there’s daily close data right? Close(0). But there is no intra-day price data?
Looking at the very best performing R2G’s from the past 360 days, I think you’d be very, very hard pressed to consistently do significantly better than their reported 1 year results. However, looking at the worst performing R2G’s (even limiting the sample to designers I know and/or feel comfortable with)…you can easily beat them by a lot.
So…Could you have picked the ‘most winning’ systems in advance? Why do you believe that? Can you predict the ‘winners’ for the next 12 months?
The central question (I think) becomes is there something you would do in building your own systems that would reliably give you much better odds at picking and allocating most heavily to only the very best systems?
Jrinne: I’m not sure that I understand the question.
If you’re asking whether the pre-built ratios use adjusted prices in valuation ratios, the answer is yes, in order to facilitate comparisons over time. Imagine a situation where a stock splits one day, pays a dividend the next day and then splits again the following day. You’d be able to compare values seamlessly regardless of which day you start your analysis. I know of no specific example that complicated, but Tootsie Roll pays regular, frequent cash and stock dividends, so it’s where I start when we test this sort of thing.
If you’re asking whether the pre-built ratios use the same subroutine as Close, I’m not sure. I think they do. I can check on that, if you’d like. The effect would be much the same, in any case.
If you’re asking whether I, personally, use Close instead of the pre-builts, the answer is no. The pre-builts use a lot of error checking code to intelligently deal with nulls in the database, differing statement periods and divide-by-zero errors. For the most part, we deal with those in user rules, but there’s less assurance. However our philosophy is that we give you the tools to calculate your own ratios. All of the valuation ratios use diluted shares, for instance, but if you want to calculate them with basic shares for some reason you can do so with your own equation.
DennyHalwes & MisterChang: I just had a discussion with a friend last week about how poor punctuation affects writing, and now I fall prey to it.
Yes, price (OHLC) and volume data is updated daily. You can expect all other data on the site to update weekly. I believe that CompuStat pushes occasional data corrections, but it’s not a normal thing as I understand it.
I’m just thinking that if the price in the value ratios are updated daily, momentum (close(x)/close(0)) and consensus earnings estimates are updated daily, a lot of ports could be rebalanced daily or on any day of the week. If I’m wrong on this, I have been pretty lucky: my Friday ports perform as well as my Monday ports and the Friday ports select different stocks.
BTW, I’m pretty sure that consensus estimates update daily for ports but not analyst recommendations. I think I’ve run screens with SetVar and watched them change from day to day. Backtesting is different.
I’m not concerned about what price information P123 uses for valuation ratios, I just wanted to confirm that it was updated daily.
I just like to be corrected in the forums rather than learn lessons in the stock market.
I have a big-time problem with the way this thread has gone and have been writing and deleting responses at various times throughout the day as I wrestle with how to reply constructively and diplomatically. But that language . . . oh my. I really need to jump in, for better or worse. So here goes.
As far as I can gather from past posts, Tom, the author of these words, is one of several on p123 who tests an enormous quantity of factors and combinations of factors in an effort to eventually settle on something that works. He and others are passionate in their belief in their process and that’s fine – for them. Any who choose this approach are certainly welcome to continue working as they wish and hopefully, p123 can continue to supply the tools and features to help them do what they do.
I strongly object, however, to anything said along these lines that is intended to or has the effect of scaring or discouraging others from developing their own systems.
The engineering approach to p123 (I’m going to use that label since it’s most ardent practitioners seem to come from that field), the stupendously labor-intensive approach about which you’ve been warned, is not the only approach. In fact, it’s not even a mainstream approach, or at least not among those who’ve made their careers in investment analysis and portfolio management, as opposed to engineering (I’ll call this the financial approach).
The financial approach is guided by ideas, not tests. It’s about formulating ideas on what you think makes for a good investment and running with those. And these ideas are actually very well-known and quite sensible. The challenge we need to overcome is to find ways to articulate those ideas in ways a computer platform can understand. Testing, rather than being the ultimate goal, is a feedback mechanism to help us see if we’re on the right track in this regard, or if we need to try another approach.
The simplest example out there is probably Greenblatt; find “good” companies and buy if you can get their shares at “good” valuations. That works. It flows directly from basic financial theory (the price you should pay must be equal to or less than the present value of all the money you expect to accrue as a result of your share ownership; good companies will lead to more money flowing to you and good valuations mean you pay less to get your hands on shares of those companies). The validity of this is every bit as inevitable as, say, the acceleration of gravity is to an engineer. Better still, you don’t need to sacrifice your full-time job, your relationship with your family, your relaxation time, etc. to get to this. It’s out there. If you haven’t already heard it, then you heard it now.
The effort part comes in how we translate good company and good valuation into computer-platform language. Greenblatt’s “Little book that beats the market” is a great starting point. It’s where he explains how he defines good company and good valuation, and better still, he tells why he defines the terms as he has. You can easily read it cover to cover without sacrificing any meaningful amount of relaxation time, family time, etc. (That book really is “little.”) Individual effort on your part will come if and to the extent you want to come up with your own definitions. That will take some work. But even so, it’s nothing to fear. It’s quite manageable; your spouse may get ticked off if you’d rather work on an idea then have dinner with your in-laws, but not to the point where you risk being served with divorce papers (unless you’re too over-the-top with the in-laws, but you wouldn’t be able to blame p123 for that – we like families). If you’re starting with a sensible idea, you’re way further than halfway to the finish line before you run your first p123 test.
Another example is Piotroski, mentioned by Tom as a model you might consider using. He’s not an engineer. He’s an accountant. And he didn’t develop his model through the engineering method. He was inspired by a great idea: He started with the well-established notion that value is good. But if life were that simple, we’d all be able to grab the lowest P/E stocks, the lowest P/B stocks, etc. and happily retire on the Caribbean island each of us would quickly be able to afford to buy. The problem is that a lot of stocks with low valuation metrics are dogs. In other words, the stocks are cheap because the companies are lousy. Piotroski got the idea that value could work if it could be applied to a subset of stocks that are not dogs. He started with low P/B stocks and the rules/filters/tests for which he’s so well-known now are those he came up with to tell computer platforms, in language they could understand, how to identify companies that really aren’t so bad. It’s a brilliant idea. It has to work. (You don’t need a gazillion tests to decide that; all you need is common sense and a basic understanding of the markets.) The challenge – and Piotroski solved it – was to find a way to express it in a manner that could be understood by computers.
So how much effort does it involve? That depends, on you; your background, what you already know, your willingness to explore new ideas, and your proficiency in expressing such ideas in p123 language). Personally, I find it an exhilarating activity that seems more like fun than work. Many others on and off p123 feel the same way (and many of who join organizations like AAII or better Investing and travel to Money Shows, Traders Expos, and the like). I’ve met many such people at events like these and each and every one of them would very strongly disagree with the dreary picture Tom painted.
That said, this isn’t for everyone. Actually, nothing is for everyone. I, for example, love strategy development, but have not been able to muster enough interest in Facebook to figure out how to actually do anything with the essentially dormant page I set up. To each their own.
A couple of other points:
MasterChang, judging by your posts here, you seem the sort who has an interest in doing it for yourself. In that case, the advice is unequivocal. Go for it!
As to whether the increased alpha is worthwhile, that’s your judgment. But just make sure you are asking the right questions. It’s not about the alphas you see in sims and backtests. A lot of that comes from dysfunctional obsessions with creating stupendous sets of test results. It’s about the alpha you show in the real world. A lot of the most spectacular R2G presentations look quite human (and sometimes less) when one limits the inquiry to live performance.
Bottom line: This isn’t about how much you work. It’s about how smart you work.
MisterChang can get Piotroski, Momentum Value, GARP and other ‘traditional’ approaches and systems from your (and other) free R2G’s or from many of the existing free P123 models. It seems to me that he’s asking a very specific different question - will he get more alpha, fairly easily from building his own systems vs. signing up for a basket of R2G’s or free P123 systems. That’s the question I was trying to offer feedback on. I’m not sure he will. Not easily. If it’s fun for him, or it’s entertainment or personal enrichment / learning. That’s all great. I like P123. Not fighting with you over the merits of the platform.
These funds are still much bigger than P123 systems ($193MM in average assets in study one, and under $300 MM in study 2).
So…I am simply saying that I think that the vast majority of that alpha can likely be captured with simpler models or a basket of R2G’s. So…how much incremental alpha can a typical investor get, above and beyond R2G’s, by doing it on their own? In 5 hours a week? If you and Kurtis and Steve and others have already invested decades learning. And you guys are offering models, will the typical DYI P123 user beat the best R2G designers after fees? That’s a valid question. I was offering my opinion on it.
P123 backtests and ‘learning’ are the best financial literacy tool I have ever used or found. I don’t find it dreary. But, if a person’s sole goal is profit maximizing…outsourcing system design may be the wiser choice.
Here’s a reason for designing your own strategy that I did not see mentioned:
I was looking at an R2G strategy today that I think could help to fill a weakness in my portfolio. So I thought about subscribing, but to me it is a black box and I would have a hard time holding it when it goes into a significant drawdown. I would rather try to develop something like it for myself so that I have a better understanding of it when it goes into a drawdown. I hate losing money and to lose money and not know why is to me intolerable. If it is my own strategy then I may be able to determine whether the drawdown is simply within the range of expectations of the strategy, or if the strategy may be flawed.
It is well known that most investors significantly underperform the very funds that they invest in! They do this by evaluating funds based on recent performance, buying what has gone up, and selling what has gone down. This is a behavioral bias that may be difficult to overcome when investing in a black box. It is also well known that most investors that actively trade underperform the market. In fact the more they trade the more they tend to underperform. But that is on average. That does not preclude a subset of the active investors from significantly outperforming, and in fact some time ago I saw some research supporting exactly that. Most active traders underperform but the best outperform, consistently. As has been said a few times already, even with tools such as P123 it is not easy to be among the best.
If you want to build your own strategy all I can suggest is to keep your expectations realistic and don’t expect it to be easy. If you want to trade someone else’s strategy, then I suggest you choose carefully and keep your expectations for the strategy realistic … and always diversify.
Years ago when I was at Value Line, I liked to piss people off by saying “we aren’t in the financial services business, we’re in the entertainment business.”
Nobody knows what true profit maximization is but those who really want to do “if” as simply as possible are just buying SPY and the line and moving on. Regardless of what any of the rest say in forums, in blogs, to interviewers, etc., they come to places like Value Line, Morningstar, AAII Stock Investor Pro, Meta Stock, p123, etc. because they WANT to. They want a better way.
What is the better way? Who knows? You can’t simply say passive with ETFs because you’re immediately take on the burden of deciding which ETFs. You can’t simply say R2G because you immediately take on the burden of figuring out which R2G(s). You can’t simply say p123 pre-defined models because you immediately take on the burden of deciding which p123 pre-defined models. And how do you know you’ve found the want to maximize profits? Yu never know. No matter what you think has been accomplished in terms of performance assessment, there’s always another week, another month, another quarter, etc.
So I prefer to skip the unaswerables and focus on what I think the real question is: in this case, whether it’s worthwhile to try to develop one’s own strategies. The questions MisterChang is asking and the statements he’s making suggest to me he has the motive to take it on, so take it on he should. As to his chances of success, I have no reason to pre-judge him or anyone else. But we all start out as novices, in everything. Everybody on p123 who has ever done everything worthwhile was, at one time, a newbie (some grew beyond that while on p123, some did before they ever got to p123, but everyone did it somehow somewhere). We have no way of knowing in advance which of today’s newbies will become tomorrow’s stars. But on p123, the tools are here, the platform is here, the data is here, and the support is here. So right at the outset, the probability of success is above 50%. Add in a does on interest and motivation, and the odds get kicked up further. So of course I’m going to encourage MisterChang to go forward.
And R2G, being what it is, always could use fresh ideas and new product. Who is to say that MisterChang, or anybody else out there, won’t be offering the top model, or one of the top models a year or so down the road.
That sort-of reminds me of the story about proposals in the early 1800s to save money by shutting down the U.S. Patent Office because all the good ideas had already been discovered. It was a bad idea back then (and one that was thankfully never followed.) It’s a bad idea now. And an analogous idea for p123 (don’t bother trying to develop your own ideas; all the ideas you need are already here) doesn’t seem any better.
In fact, a textbook project on which I’m working with a professor who built a graduate-level course around p123 is opening my eyes to entirely new set of ideas, to the point where I’ve lately come to think p123 has barely begun to scratch the surface of what’s achievable. So . . .
A lot. A helluva lot. And far far more than some realize (and hopefully, some of the stars of tomortrow will bring some good new things to R2G).
@ Marc
I disagree with your quote related to mass testing rather than developing ideas. I exchanged often emails with Tomyani and while it’s true we might have a huge load of tests, each test came from ideas we either invented ourselves or read on SSRN, newspapers, articles, etc. There is more time reading and thinking than time spent spitting test results. Therefore it’s not really senseless amount of permutations and feeding a huge monster. But I understand what you mean about an engineering approach to a problem of non-financial background who comes on P123 where it’s mostly about constraints and goal seeking answers.
