Assume for a moment that you have a black bag containing two balls, one red and one black. You pluck a ball out of the bag and hold it tight in your hand without observing its color. The odds that the ball in your hand is red are 50%.
Now you look in the bag and see a black ball. The odds that the ball in your hand are red just increased to 100%. Nothing changed, except new information.
Here we have an example of an underlying truth (the color of the ball in your hand) that we cannot see directly, yet by piecing together various data, we can determine with certainty the color of the ball in your hand. You were given information about the environment from which you selected the ball. You then peered into the bag to observe its contents. The combination of the two pieces of data gave you certainty about the color of the ball in your hand without observing it directly.
If a patient tests positive for AIDS, what are the chances that the patient actually has AIDS? This is the same type of problem.
Any random patient that tests positive for AIDS has less than a 2% probability of having the disease. That number is low because the prevalence of the disease among the general population is very low and the test is imperfect (we assume here a test that is 95% accurate). The combination of even a slightly inaccurate test and a low incidence rate means the test results in lots of false positives. Those false positives overwhelm the true positives.
If we determine the patient is male, the patient’s chance of having the disease goes up to about 12%. Why? Because the prevalence of the disease among men is higher than for women. There are still lots of false positives, but by hiving off a portion of the reference class (the women) with super-low incidence rates, we are left with a reference class with a far higher concentration of patients with the disease. The higher incidence rate means fewer false positives relative to true positives, which in turn means a patient with a positive test is more likely to have the disease.
If the patient is a gay man, the probability increases to 77%. Nothing has changed with the patient or the test; the exact same patient took the exact same test with the exact same result. In obtaining the new information, we hived off another large portion of the population with a low incidence rate, which left us with a reference class with an even higher incidence rate. The higher incidence rate of the smaller reference class reduces the number of false positives relative to true positives even further.
By successively narrowing the reference class, we zero in on the underlying truth of whether the patient has AIDS. In the 2% case, the patient came from the general population. In the 12% case, the patient came from a portion of the general population that contained higher concentrations of people infected with AIDS. In the 77% case, we narrowed the reference class still further to a portion of the population with even higher incidence rates.
Most gay men that test positive for AIDS then have a second type of test. Virtually 100% of gay men that test positive with both types of tests actually have AIDS. With the second test, we reduced the reference class to such an extent that we eliminated all false positives entirely.
We call them proof points. Proof points are little pieces of information that reduce the reference class in a way that reduces the number of false positives relative to true positives. Observing that black ball in the bag is a proof point as is asking a patient that has tested positive for AIDS if he is gay. Proof points allow us to zero in on the underlying truth.
As investors, we rarely see the underlying truth directly. We get clues and hints but cannot access the underlying truth itself. When we analyze a company’s competitive advantage, for example, we never know with certainty that we are right. We deal with uncertainty and probabilities.
If we randomly pick a company from the general population of companies, the odds of that company having a strong competitive advantage are miniscule. The prevalence of strong competitive advantages among the general population of companies is tiny.
But we don’t randomly select companies from the population at large. We look for companies with certain characteristics. We use certain metrics.
We believe dLocal offers a unique product and has a strong competitive advantage built on what we call local scale. If we picked dLocal randomly from the population of all publicly traded companies, the odds that it would have a strong competitive advantage would be very low. Almost nonexistent. But we don’t pick companies randomly. dLocal has certain characteristics we like. We put it through our evaluation process. We liked what we saw.
Armed with only the results of our evaluation process, the chances of dLocal having a strong competitive advantage are still quite low. We like our evaluation process, but even the best process applied to a population of companies with a very low incidence rate is bound to come up with some false positives. A false positive in this case is an “imposter” that talks a great game and exhibits good metrics but has a lousy or nonexistent advantage. We want to do what we can to keep imposters out of our portfolio.
That’s where proof points come in. Proof points help us shrink the reference class in order to maximize the number of true positives relative to false positives that make it through our research process. One type of proof point we love are high-stakes decisions made by vested and knowledgeable participants in the value chain.
If a top brain surgeon had to have bypass heart surgery, anyone looking for a good heart surgeon should take note. The brain surgeon supposedly knows something about surgery and is clearly a knowledgeable and vested decision maker. The reference class from which the brain surgeon selects his doctors is likely to have very few if any false positives. The decision made by the brain surgeon, in other words, would be a wonderful proof point for anyone looking for good heart surgeons. It reduces the reference class in a way that almost eliminates false positives.
To qualify as a proof point, the person making the decision must have a significant life-altering vested interest in the outcome of the decision and must have “inside” knowledge about the technology or value proposition in question. The combination of a life-altering vested interest and inside knowledge makes for a powerful proof point.
When Pedro Arnt left Mercado Libre for dLocal in August of 2023, he was a knowledgeable participant making a life-altering decision. He had been with Mercado Libre almost from the beginning, ultimately becoming its CFO. Mercado Libre is today the largest e-commerce company in Latin America. As its CFO, Pedro had intimate knowledge about payment systems in emerging markets and importantly, dLocal’s value proposition is facilitating payment processes in emerging markets. At Mercado Libre, Pedro saw dLocal’s capabilities firsthand and from the customer’s perspective. He made a life-altering career decision based on what was likely “inside” information. Bingo.
We never spoke to Pedro about dLocal’s competitive advantage. We didn’t have to. His decision, which was public for all to see, told us everything we needed to know.
Most asset managers fail to fully understand the effect of false positives. No research process is perfect. Even the best processes result in a false positive or two. False positives are the silent killer of performance.
Asset managers that ignore the effects of false positives also fail to fully grasp the value of proof points. Proof points are the little gems that prune the reference class in a way that eliminates or reduces false positives. Pedro’s career decision was like the brain surgeon selecting his heart surgeon. Pedro made a life-altering decision based on inside information. Most investors ignored the move. We didn’t.
