Hi All,
I bet I got your attention with the “secret sauce” bit.
I am posting with the hope of getting a thread going that will help others in their quest for “above average returns”. What I have noticed is that with the advent of R2G/Smart Alpha/DM many developers have become guarded and less willing to share their findings. I am guessing the great pull of monetization to be the cause. Just go back in the old forums and read the likes of Denny”DennyHalwes” or Oliver”olikea”(my apologies to the many others who have also made significant contributions) and you will get a sense of what I am talking about. Anyone who is serious about this game should go back and read all of their posts. It will take you some time but there is a wealth of knowledge there.
Here is what I propose. Anyone who has benefited from P123 I ask you to share some significant finding that has allowed you to find success. In essence I am asking you to “Pay it Forward”. Now I am not suggesting that you give away the farm, you don’t have to give specific parameter sets, what I am saying is that you give others(who are willing to do the work) an idea or a way forward so that they can develop their own findings and share them with others in return.
One more thing, NO HATERS PLEASE. I want this to be a thread where people share without fear of prosecution. People are less likely to share if they are getting chastised in the process. If you don’t agree with what they say, fine, just move on to another thread or better yet share your own successful findings.
This may fall on deaf ears but I thought I would try to get some real insights flowing again. So who is willing to give something back?
Diversification- some of which you may have overlooked, the simplicity is powerful
Diversification in the total number of stocks. Each person’s risk profile is different but 40 stocks is my minimum. At 2.5% per stock I don’t worry about the 50% hit on any one stock.
Diversification in Portfolios. I like to use a book of at least 4 portfolios(ex. Value + Momentum, Sector Momentum + Value, Strict Value, Growth +Momentum are some of my favorites)
Diversification in Ranking systems(there are many good ones that are available free on P123)
Diversification in Ranking system factors. I have found that it is better to have more factors rather than less. But they do have to make sense(I am with Marc Gerstein on this one) to avoid the curve fitting bias.
Diversification in Market Cap. I do have a bias for small Caps but they don’t make up 100% of my book.
Diversification in Market Sectors. I do have a bias for the sectors that have the strongest momentum but they don’t make up 100% of my book.
Diversification in market timing. It is not necessary to be all in or all out. As the market goes down my risk exposure goes down. As the market goes up my risk exposure goes up. All in or all out leads to whipsaws that are hard to take.
The overall effect of all this diversification is a much smoother equity curve. This won’t improve your top line that much but it will sure protect your downside. This helps me sleep at night. When I first started trading I remember something that an oldtimer said to me; “in the markets any fool can make money but very few can keep it”.
Hope this helps. Who’s next?
It’s not much, but I’ve had some luck dividing my universes into dividend paying stocks versus non-dividend payers. I have found that developing ranking systems and port rules for each and then combining them into a book has worked far better than trying to develop a single port that handles both.
I have found that low liquidity stocks are a significant source of alpha. Trading them live is not nearly as bad as it seems and you get paid very well for the work of managing slippage. I have found that the profit outweighs the cost of getting stuck in the occasional liquidity trap.
Finding things that work in both Canadian and US universes is a great source of comfort when you’re worried about curve fitting (which I think we all need to be).
I think it’s fair to ask for ideas, and which ideas have worked so far. Instead of re-inventing the wheel, I based my models on well known investing books (Greenblatt, Piotrosky, Graham, Lynch, Fisher, ONeil) and on white papers, which was done with way more research and deepness that I could have ever implemented (Fama & French, Andy Cooper, and others from SSRN and AAII - tons of reading material there).
My setup includes Canadian and US portfolios. These markets are fundamentally different - my Canadian models do poorly in US Exchanges and vice-versa. I have the following live models today:
A model oriented towards income and quality: A company that has a low beta historically is not typically volatile; therefore, if its current yield is higher than its historical yield, a case can be placed for valuation. When combined with metrics to identify quality, it can be used to find companies that have fundamentals disconnected from price. Lastly, momentum rules can be used to detect when a potential reversal took place, since companies that have lost the love of the market can remain undervalued for a long time.
A model oriented on momentum, which is basically what we observe (after the fact). Technical Analysis can be used to identify patterns to predict an outcome (as described by ONeil in “How to Make Money in Stocks”). This takes advantage of overvalued market that continues to produce momentum, identifying stocks that my value models couldn’t find.
A model focused on value, sentiment and momentum: If a stock is undervalued, sentiment is positive for growth and it has caught momentum, it’s likely to continue to outperform. Inspired on this paper: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2895101
All models use market timing, based on a combination of factors involving fundamentals (with technical oscillators for SPCNY) plus technical analysis (MACD and EMA crossover on benchmark price) plus economic indicators like UNEMP rate. Fred Piard has some great articles about this.
Regarding ETFs, George Vrba also has some interesting models oriented in outperform specific ETFs.
Each of my model is designed to hold between 7 to 10 positions, so when combined they provide a decent diversification.
I personally find a lot harder to design a trading model than to invest, because trading requires 2 luxuries not needed when investing: Locking profits for the short term and reducing drawdown.
Besides reading olikea and danny’s post, I’ll add to read all Marc Gerstein posts too. Hopefully this will inspire others with ideas to choose the best apples instead of getting the whole basket.
In my work (10 years on here) the biggest breakthrough came when Variables were introduced. This allowed creating screens programmatically, to a point.
When variable position weighting is introduced it will be a game changer…
thanks for sharing your insights. can you elaborate a bit more on managing slippage of low liquidity stocks?
I have tried some time ago and the sims looked great but during real life trading I found the slippage too much. Of course it could be that very case that I used market order on the opening or close.
I’m very new to p123, but some learnings or thoughts. (I’m still in process of building things out, so take all of this with a grain of salt.)
Evaluate your investing personality and build models that match your activity level/desired rebalance period. Purely by the numbers it often makes sense to become a trader and crank the systems down to weekly rebalances and go for starry-eyed returns (and I’m sure some people are getting them), but in real life I’m probably not suited for that much trading activity on a regular basis, at least not yet. Record-keeping, logging into all the brokerages, managing the trades. I end up spending entire days watching the markets if I’m trading a lot. So I’m focusing on longer term holds now to suit my disposition. I think I’m the type of investor for whom fewer decisions and points of action make a better real life system. I absolutely enjoy building the systems, but actually actively dislike implementing the actual trades unless it’s a simple high liquidity/tight spread/easy transaction. Unfortunately, the better opportunities usually aren’t highly liquid trades with tight spreads. I need to get better at that, especially studying trading more illiquid issues, but in the meantime I can help myself by building compatible models.
This is not P123 specific, but I’m probably more risk averse in reality than I think I am in an abstract sense, so I pay particular attention to drawdowns vs. the benchmark, sharpe ratio, standard deviation, and just generally a smoother looking curve. My hypothesis is those 50% drawdowns don’t look as big in a backtest as they are in reality.
If you’re a little guy, and don’t need to trade large amounts, and can have long holding periods, and can be patient buying/selling, it seems the data supports that there’s a sizeable illiquidity premium, and this is likely a space a bit insulated from funds that have much larger $ they have to deploy. Be cognizant of difficulty getting in and out of positions (hence preference for long holding periods), slippage, spread, and impact on price - but it seems the little guy has opportunity here based on the numbers. Interestingly also, these types of companies/stocks in a system seem to be less volatile also, assuming you can ride things out, so there can be a diversifying benefit. With P123 it’s not difficult, even putting in large slippage, to find sizable illiquidity premiums. 6 month rebalance periods seem viable to me. I have that model on the to do list.
Use EvenID=1 and EvenID=0 as lines in your universe or screen to create a test and control group. I only learned about it a short while ago, but it’s a great quick way to test your finished system against data that was not utilized in creating it.
Even a pretty weak, but uncorrelated independent factor, can help improve your model. Just because you test something and see little advantage, don’t immediately discard it if it’s something that makes sense as probable source of mispricing. It might work well in combination with something else. I was shocked the first time I saw this happen - I think it might’ve been accruals. And similarly, two independently really strong but correlated factors sometimes don’t help at all, and can hurt when combined.
And finally for a more speculative thought. As I’ve been working on ranking systems, I’ve begun wondering if possibly there’s some advantage hiding in the NAs that we typically discard. You know sometimes when you see that little bump up at the far left side of the rank distribution where the NAs get sorted? Are there good companies in there that are more likely to get missed because of missing data or not calculable, and can they be sorted? I don’t know, but I’m thinking maybe?
Anyhow, again I’m just learning, but maybe it’ll be helpful to somebody who reads,
Yes, yes, yes. And never feel apologetic for selecting longer rebalance periods. Certain kinds of ideas-stories need more time to play out than others. Some strategies work better with weekly refresh. Others work better with four or 13 weeks.
Perhaps the worst lingering consequence fro 2008 in a p123-user context is that it distorted expectations. That was a generational financial system crash. There was no place to hide and actually, the strongest best stocks often performed worst because those were the ones for which liquidating funds could most easily get bids (and a lot of p123 forum conversation back then discussed this as “rank inversion”). Many think they can model away from similar crashes in the future, but I would not be confident in any of those approaches. Every major crash has its own unique causes and characteristics. So while its as easy as easy can be to create timing systems after the fact that could have gotten you out ahead of 2008, don;the assume they’d work in the next one. The best protections against this sort of pain are being ware of what’s going on in the world, and broader financial planning (i.e. managing how much of your total capital you want to expose to different levels of equity risk). Also, unless we go over a permanent or very very very long-term cliff, those mega-drawdowns tend to be followed by mega draw-ups on the other side, and those too come suddenly and many who try to model ahead miss them.
Yes again, and many on p123 have been able to benefit from this, although many have done it with weekly models. I haven;the been in this area of the market lately so you may want to start a separate thread seeking discussion with others who are more up to date than I am, but from my observation, easy money (accommodation by the Fed) has enabled money to move comfortably around in this area so long as nobody gets out of line with how much of a position they want to move at a single point in time. Even under the best of conditions, you’re at risk of news opening up a monster spread here and there, but I sense that those here who are active in illiquids have been able to cope. Anyway, you might attract more knowledgeable responses from others from a new thread with a new title.
On paper, that sounds great and you can use even and off numbers for market cap, sales or anything else. But in terms of ID, I’m not sure it is as clean as it should be. I’ve often had models that weren’t so great in this regard, but which have been very successful for me in real life with real money. I’m not enough of a statistical theories to delve into the whys and wherefores.
Yes, no stock moves based on any one thing. If I’m a value investor, it makes sense that I’d be better off when I want to sell being in a stock that could motivate dean not just from other value investors, but, say, from growth investors too. Never be reluctant to sell to and take profits from other market participants whose ideas you disagree with.
Separately, even the addition of highly correlated factors can be a plus although pure statistics would suggest otherwise. What pure statistics misses is the potential for noise in any given data set. Having more ways than one to express the same broad idea helps you diversify against such aberrations. We’re used to a portfolio of stocks. We should similarly be wiling to think in terms of a portfolio of value factors, a portfolio of growth factors, etc. Put another way, the MPT (Modern Portfolio Theory) concepts of systematic and unsystematic risk apply equally to portfolios of investing factors within the same style only here, the risk is the presence of information that can lead you astray.
Actually, there is a choice you can make to push NAs to the neutral position.
[quote]
Perhaps the worst lingering consequence fro 2008 in a p123-user context is that it distorted expectations…
[/quote]Yes. To quote Peter Lynch: Far more money has been lost by investors preparing for corrections, or trying to anticipate corrections, than has been lost in corrections…
[quote]
Many think they can model away from similar crashes in the future, but I would not be confident in any of those approaches. Every major crash has its own unique causes and characteristics.
[/quote]The data backs up this assertion. There was little correlation between the factors that did well in the Dot Com crash to those that did well in the Financial Crisis.
[quote]
So while its as easy as easy can be to create timing systems after the fact that could have gotten you out ahead of 2008, don;the assume they’d work in the next one.
[/quote]This applies to ETF systems too. I have designed ETF systems with > 45% a year returns in backtests, but I don’t use them and don’t market them. ETF systems are far less robust. Compare the actual performance of well designed ETF R2Gs vs. well designed stock R2Gs.
[quote]
…unless we go over a permanent or very very very long-term cliff…
[/quote]I posted at length about these a while ago. Long term cliffs have happened around the world. Study them to recognize them and protect yourself.
2008 was a systemwide crisis that impacted pretty much all financial assets.
The Dot Com crash was much, much different. It was primarily the unraveling of a narrowly focused valuation bubble. Many stocks did relatively well. And it’s a shame p123 wasn’t the born until 2003. Had it been founded five years earlier, many, perhaps most, users would have sailed through the crash because there was so little appreciation of the sort of fundamentals we use. That’s why I don’t run tests or sims that far back unless I have to do it for publication, etc. When I evaluate for myself, I don’t the want to look at that period because it was so exceptionally easy for us to outperform by wide margins.
1998 was a less intense version of what we saw in 2008.
1994 was an emerging currency crisis.
The Gulf war was a more classic recessionary thing.
1987 was a valuation and trade deficit excess.
1981 was excess in inflation-interest rates and that may be the most important one to this day because business,political and economic leaders all came of age at that time and that’s why they are scared to death to let interest rates and inflation move up from today’s way-too-low levels – seriously, it;s been at least a decade and likely much more since any company shelved a project because cost of capital was too high).
They’re alway different. But the one thing they all have in common is that they come about in response to some sort of “excess.” I think it would be very helpful for folks here to set up stored collections of economic charts and pursue them periodically to look for things that seem to be moving to excess. The one catching my eye right now is velocity of M2. That comes from the things that caused the election to go as it did in 2016 – and it’s continued to deteriorate since then. (This is a likely consequence of the above-mentioned inflation paranoia).
Even if you leave out tech stocks, the largest companies in the other sectors vastly underperformed the average sized S&P stock during the 2000-2003 crash. Normally, larger companies are a safety play in a bear market. But 2000-2003 was different.
This is because the mega cap stocks vastly outperformed in mid and small cap stocks from 1995-99 (wouldn’t you know it - right after Fama published about the size anomaly in 1994. Bad beat, Gene). Generally, what goes up fastest, comes down fastest. From 1/1/95 to 1/1/00:
S&P 100 Index = 30% per annum
S&P 500 Index =26% p.a.
S&P 400 Midcap Index =21% p.a.
S&P 600 Small Cap Index =16% p.a.
Russell 2000 Small Cap Index =15% p.a.
Dot com played a role in those differences, but did not account for all of it. For example, the largest quintile of the Dow crushed the smallest quintile of the Dow from 1995-2000.
This is why many stocks did relatively well during 2000-2003 – they had trailed for so long that they were bargains. 2000-2003 was just a big mean reversion compared to what happened from 1995-2000.
Sorry, but I didn’t mean for my comment on sharpe ratio, drawdown, volatility,performance smoothness, etc to mean market timing approaches. Most (maybe all?) systems I’m looking at are just simple fully invested at the moment. I should’ve been more clear that I’m looking at fully invested system characteristics, not timing/hedging. I find particularly valuable the rolling backtest summaries and love it if I can find a model that tends to or exceed the market in up periods, and fall only about half as much in down periods (looking a 4wk,8wk,3mo,and 6mo holding periods these tendencies tend to reveal themselves). It may be folly to think it’ll hold up in the next bear with unknown origins, but I’m really trying do things that help counteract the fear and urge to “get out” that can set in.
Where the correlation observation hit me really hard: I recently spent a good while looking at growth factors where the middle of the distribution is where the mispricing seems to occur, and no matter what I did (looked at at all manner of sales gro, bvps gro, op inc gro, eps change), it seems there was very little benefit I could add to what sales gro does independently. What I noticed happening was as I kept piling more similar factors on top of each other the spread between the top50 / bottom50 would usually increase (which would be great if I was buying half the market), but there are often not benefits at the top of the distribution that I’d be ranking stocks by. I think I’ve noticed a similar dynamic with value factors - after 2 or 3 of them it seems like marginal gains are small, and can hurt. But like you say, even if it hurts the backtest, maybe it helps prospectively. James O’Shaugnessy advocates lots of similar factors because some come and go out of style on historical time frames and can stay out of favor for long periods, and we really shouldn’t just bet on the best ones. Wes Gray I think advocates more for picking the single best factor for measuring an idea and sticking with it. Overall, it seems results for the 17-18 year window we have usually favor combining more factors up to a point (I do think it can be overdone but throwing the kitchen sink does seem to be surprisingly effective). The growth factor (at least the way I was looking at it) was a tough one, that defied that expectation. Maybe I need to take another angle on it.
I was young then, but do remember thinking a the time that Blue Chips looked expensive. Didn’t realize it was that kind of a run. As a Buffett fan in training I specifically recall looking at KO at the time and wondering why people were buying. If not for the taxes I suspect he would’ve been selling.
I haven’t said too much in over ten years on P123 but your “secret sauce” request got my attention. I have read the previous replies with interest and agreement. I will just list the things that seem to make P123 work for me.
-First, I started out in the era you describe when other P123 members were more free with their ideas and learned much from the people you mentioned and others. There were many ranking systems developed that I used or revised or got ideas from. I really could never develop ranking systems that worked any better than the ones that have been made public on this site. Of course I started out trying to hit home runs with small cap systems and developed some that worked that I still use today. The ranking systems I like that are available are Advanced Momentum Value, Bompusrank, and variations of TF-12, and also Balanced4 is a good ranking system.
-Second, I joined P123 to invest my company account and 401k account myself. I had a good broker, but he was managing 100+ accounts besides mine. I thought I could do better. I quickly found I needed large cap stocks or etfs to feel comfortable putting a large portion of my savings into the market.
-The biggest thing on P123 that has made a difference for me is Books. Which allowed me to mix different systems together. And as Marc says, you can’t model away the last crash. The next one is coming and it will be different. So the best you can do is diversify with non-correlated ports/assets. P123 books has allowed me to put together some of my ports with other ports or etf’s that offer really good risk/reward profiles. It’s the only way I invest now. If you are new to P123 you can go pick 3 to 5 designer models and come up with a good risk profile that will beat the S&P. Heck the best performing large cap port is Marc’s “Cherrypicking the Bluechips” and it is still free. I don’t want to make this sound easy because it is not. It is hard and requires work but P123 adding Books was the difference for me.
-As far as “Secret Sauce”, one thing that I have done that has worked for me is using a timing system that works directly on the equity curve of the S&P. Can’t really do it like I like in P123 so I use an excel program and enter it in the exposure list. It is based on technical signals (really just MA’s) and variable time frames depending on market action (up/down) and does not try to micromanage your port, just gets you out of the big down drafts and will get you back in at a reasonable point. We will see how well it works in the coming crash. Might have started today.
That’s about it: Books, Non-correlated assets, Timing (loose) helps and keep your expectations in check.
Indeed. I always check factors for how many N/As they produce. If they produce a lot of them, there are workarounds. There are some factors here that typically have 30% N/As but that one can calculate in another way. I also use conditional nodes so that if, say, a company has CurQEPSMean = N/A I can use EPSExclXorQ instead. I also only use “Percentile NAs Neutral” for all my ranking systems.
Here are ten of the ingredients in my secret sauce, in no particular order. I’ve taken them from my blog, www.backland.typepad.com/investigations, where you can read a lot more about each of these tips.
• There are five main things I look at when I invest in a company: growth, value, quality, investor sentiment, and size (smaller is better).
• If I can rank stocks on a factor, I don’t screen them on that factor.
• The bigger the company, the more efficiently its stock is going to be priced. The only way I can make alpha is by exploiting market inefficiencies, and there aren’t as many in the large-cap space.
• So-called price trends in stocks—at least those that are measured in days or weeks—are simply an optical illusion. A “trend” is more likely to reverse than to continue. Since trends don’t exist, I don’t have to watch for them. I don’t have to wait until a stock is “finished trending upward” before I sell it, and I don’t have to wait until it’s “finished trending downward” before I buy it. I don’t have to use trailing stops or look at daily price fluctuations. I know that any direction a stock moves in is more likely to reverse than to persist. So I buy low, wait, and sell high.
• The best way to find out whether a backtesting procedure will produce strong results is to do correlation studies. Develop fifty different strategies and see how well their simulated results correlate over many different time periods using different procedures. For example, a five-stock simulation will probably be less correlative than a fifty-stock simulation. A two-year simulation will probably be less correlative than an eight-year simulation. Alpha will probably be more correlative to forward returns than the information ratio. And so on. The best backtesting procedure will be the one that produces the most correlative results.
• I invest mostly in microcaps—under-the-radar companies. I have a portfolio of fifteen to twenty-five stocks, and I hold my investments for an average of two to five months. I’m looking for companies with the following characteristics, among others: a comparatively low stock price, healthy free cash flow, middling but accelerating revenue growth, low and stable accruals, low trading volume, and comparatively high earnings growth. In addition, I avoid companies with very low share liquidity or whose stock price is below a dollar or two per share, master limited partnerships, real estate investment trusts, utilities, and companies from countries that rank high on corruption indexes.
• If I’m going to spend tens of thousands of dollars on something, that something better be good. I treat every investment like a precious object, evaluating it from forty different angles before I buy it, and selling it, usually at a profit, only when I need the cash to buy an even more precious object.
• I engage in robust backtesting, varying my tests over different time periods, with different numbers of holdings, over different holding durations, and using different stock universes.
• The only way to invest profitably is to implement a strategy that you’re completely sure will work and stick with it, making incremental improvements to avoid too much buying and selling.
• Certain financial metrics are so widely used that they can give me no advantage over other investors: price to book value, for instance, or price to trailing-twelve-month earnings, or EV to EBITDA, or return on equity. I don’t use those. There are still plenty of financial metrics that many investors don’t pay much attention to but make a big difference to a company’s future performance. These give investors like me an edge.
I need to get better at that, especially studying trading more illiquid issues, but in the meantime I can help myself by building compatible models.