Stop worrying about your edge
It's fucking pointless
Investing is often compared to poker, painting it as a game where it is possible to size up all opponents before deciding to play. “If you can’t spot the sucker in your first half hour at the table then you are the sucker” is a quote that I even saw Buffett paraphrase. Problem is though, there is no way of telling why the person or institution at the other side of a transaction is buying or selling. At least I haven’t yet been able to find this functionality in my Interactive Brokers trader workstation. When I look at a price graph of a stock, or at an order book, I have basically zero information on what the other market participants are like. They could be someone’s cat walking over their keyboard for all I know.
I may read someone’s take on Twitter to try and determine my edge before buying a stock, but there is no way to know if their views are representative. Maybe the stock has traded down because nobody paid attention and most market participants are day traders? Maybe there was a forced seller? Or maybe hundreds of very bright investors with large trading accounts looked at it closely and passed? Not even showing up in the order book. And none of them may have a Twitter account (after all, they are very bright).
It can be quite counter productive to worry about having an edge in investing. I have seen intelligent people only focus on overly complicated situations, thinking that their 1000 IQ galaxy brain could provide them an edge. These people probably look at relatively simple mispriced situations and think that they must miss something and move on. My article on the Michigan State lottery was a good example of an obvious and straight forward edge that went unnoticed for years.
But to understand why the comparison does not make sense we have to take a closer look at the mechanics of poker and zero sum games in general. And then try to apply that to investing.
In poker it is quite complicated to get close to solving the game. It is basically impossible for humans since the game is too complex. In a two person symmetric zero sum game, solving it means finding the one unbeatable strategy, the Nash equilibrium. Where even if your opponent solves the game as well (which means they discovered the same strategy), the best both of you can do is break even in the long run. In every such game (which investing basically is, assuming only a small percentage of volume is insider trading) there is one and only one such strategy.
So the process of “finding the sucker” in practice means “finding the player whose playstyle deviates too much from the Nash equilibrium”. And this sucker is exploited by taking the opposite line, or by simply following the Nash equilibrium strategy. The latter being a lot less risky, but also less profitable. To illustrate this with an example: if someone bluffs too much you can start calling lighter than what is optimal to get a higher win rate, but then you also risk being exploited if your opponent adjusts by then lowering their bluffing frequency enough. The Nash equilibrium is found when neither side can make profitable adjustments anymore.
When both players have solved the game, they know exactly their opponents hand range in any given situation, but cannot use this information to win, only to break even at best. Minus any rake that is taken out by a third party.
Because this is so difficult for humans to do, it is best to play really bad players so you can be sure you have a better approximation of the Nash equilibrium than your opponent. In 2017 a poker bot was created to basically solve heads up No Limit Hold'em and it demolished top players by a pretty wide margin. But it still took some pretty powerful computers months using a regret minimization algorithm, for only one stack size of 100 big blinds. Taking up 2.6 petabytes of storage during the competition.
A way to make this concept more intuitive is to take a simpler game like Rock Paper Scissors instead. Solving this game would be very easy due to its low number of game states. The solution is to randomly pick Rock, Paper or Scissors ⅓ of the time each.
The difference between Rock paper scissors and Poker is that in poker the Nash equilibrium strategy is actually profitable. Whereas in RPS you only guarantee a break-even EV outcome against any other strategy because it has a mixed Nash equilibrium.
Let's do a thought experiment and see what it would take to solve value investing. If it is even possible. A bot that purely focuses on correctly valuing a company's cash flow stream, that tries to predict how productively it will be reinvested and what % of it will actually reach minority investors. (So not a trading bot that finds patterns in market movements).
The goal would essentially be to predict which stocks will generate and pay out more or less free cash flow than the market is currently pricing it at. Or more accurately assess the outcome of special situations. Better than humans can. This bot would obviously have to be pretty good at finding and interpreting messy data. And have to be good at natural language processing. Other than looking at market prices and insider transactions, it would pretty much completely ignore other market participants.
We run into some problems here though, because to solve a zero sum game using a computer, a clear and easily verifiable outcome is needed. Which is easy enough in poker: either a hand wins or loses according to its static rules and a computer can easily simulate a large amount of representative situations per second. Based on this outcome, the computer can adjust its strategy over and over playing against itself, until no more profitable adjustments are possible.
But how to simulate the amount of profit a company will generate? And what is done with those profits? As companies do not exist in a vacuum, the bot would have to make accurate predictions several decades into the future about micro and macro economics, technological breakthroughs and consumer tastes. Poker involves only the cards on the table and the players around it while investing really involves almost everything happening on our planet to some extend.
You could do it with historical data, but then you run into the problem that the rules of the game were drastically different 50-100 years ago. If you create a bot based on even 30 year old data, it might show that local newspapers are a good wide moat investment, not factoring in the existence of the internet. Plus the amount of data would obviously not be even close to all the possible meaningfully distinctive outcomes.
Finally even if we could somehow isolate a portion of reality and simulate it, the amount of game states would be too large and can probably not be pruned effectively enough like in poker. Since solving a zero sum game is essentially done through intelligent brute forcing, the larger a game gets the harder it is to solve it. Libratus didn’t even entirely solve NL Holdem (within about 5BB/100 hands by a rough estimate of one of its creators).
Counter intuitively though, the difficulties of making a bot to solve a zero sum game can make it easier for a human to have an edge. Take for example limit holdem, due to its fixed size betting, it is significantly less complex than no limit holdem. And a close enough approximation of the Nash equilibrium was found much earlier. This is relevant because you want to play against other flawed humans who are quite inefficient. But as soon as the blueprint for a superhuman bot is created, most of the opposition will be bots as they can be quickly replicated, making the game hard to beat for humans.
So until we get AGI’s that can be cheaply replicated, value investing will likely stay profitable for humans. Which will probably be at least another couple of decades?
We can learn some lessons from poker though. Good poker players don’t usually put their opponents on single hands. They put them on hand ranges. This is comparable to coming up with a valuation range for companies instead of a single price target. The higher your ignorance, the wider this valuation range and the more smeared out the probability distribution will be. For the average pre revenue biotech it will be very wide, and for Pepsico it will be much more narrow.
Obviously having the discipline to wait for a juicy game full of whales is another similarity. This is probably the hardest part, and something I still screw up semi regularly. You want to only invest in stocks that keep you awake at night because you have not bought them yet. I started to see the value of always holding some % of cash for this reason as it keeps me looking for cheaper stocks.
Finally the biggest mistake I often see is focussing on outcome instead of process. In poker if you make a bluff and get called, you don’t judge success or failure by outcome, but by whether the play itself was sound and was profitable against the opponent's estimated hand range. Now obviously if you do that specific bluff a thousand times and you end up losing a disproportionate amount of time, something is wrong. But never judge the soundness of a bet based on a single or a handful of outcomes. Or on information that was only known in hindsight. This is called process oriented thinking vs outcome oriented thinking. And it is actually quite difficult for humans to do naturally.
So if a stock goes up further after selling it was not always a bad idea to sell. Because if it is then what game are you really playing? Are you momentum investing or value investing? When your mantra is to sell losers and buy winners, you are not value investing. With momentum investing and technical analysis you do very much have to worry about other market participants. Which has made far fewer people obscenely wealthy in the past 100+ years. And you are probably competing with trading bots from Renaissance Technologies.
So does this mean edge is not important in investing? Well it is actually. But there is no way to know if you actually have one. You just turn over a lot of proverbial rocks and place your bids or hit those asks and pray to every deity out there your logic is more sound than the person on the other side and you did not miss something important. Or that you are buying from their cat walking over their keyboard accidentally liquidating their account with a market order.
PS: for anyone who wishes to test their poker skills, there is Slumbot, a free browser poker bot that probably beats most professional players, although not as good as Libratus. Stack size of 200 big blinds. Playing around with this, it is interesting how it would do things that are typically associated with bad players, like “donk betting” the turn after calling the flop. Or regularly check raising a middle pair on the flop. Showing how the GTO way of playing a game sometimes means making some pretty counterintuitive moves.
PSS: If this article gets at least 10 likes I will write about a new useful mental model I have found. I have to get some of these more philosophical ideas out of my system by writing them down.
Loved it.
I believe one edge a value investor can know he has is discipline.
It is the hardest one to get AND to keep, though.
Good stuff! Looking forward to more of these