How Tails Constrain Investment Allocations

How low probability-events restrict advisable position sizes

You would need to be living under a rock to not know about the importance of small probabilities on asset distributions. By 2020, every investor has been Talebed to death by his golden hammer. But knowing and understanding are not the same. I know it’s painful to give birth. But if I claimed more than that I’d end up only understanding what it felt like to be slapped in the face.

I’m hoping the above discussion of the devilish nature of small probabilities makes the seemingly academic topic of fat-tails more visceral. But if it didn’t I’m going to try to drive it home in the context of a real-life investing decision.

Step 1: Understand the impact of fat tails

I ran a simple monte carlo assuming the SPX has a 7% annual return (or “drift” if you prefer to sound annoying). I assume a 16% annual vol or standard deviation and ran a lognormal process since we care about geometric returns. We’ll call this model the “naive simulation”. It does not have fat tails.

Based on these parameters, if you invest on January 1st:

  • You have a 5% chance of being down 23% at some point during the year.
  • You have a 50% chance of being down 7% at some point during the year.

Now be careful. These are not peak-to-trough drawdowns. They are actually a subset of drawdown since they are measured only with respect to your Jan 1st allocation. The chance of experiencing peak-to-trough drawdown of those sizes is actually higher, but these are the chances of your account being X% in the red.

That’s the naive simulation. To estimate the odds in a fat-tailed distribution we can turn to the options market which implies negative skewness and excess kurtosis (ie fat tails). I used 1-year option prices on SPY. Option prices answer the question, “what are the chances of expiring at different prices?” not “what are the chance of returning X at any point in the next year?”. To estimate what we want we will need to use the pricing from strikes that correspond to the equivalent one-touch option. Walking through that is overkill for this purpose but hit me offline if you want to see how I kluged it.

Let’s cut to the market-implied odds.

  • You have a 5% chance of being down 39% at some point during the year.
  • You have a 50% chance of being down 11% during the year.

Now you can see the impact of fat-tails: the gap between 23% and 39%. This is the impact of kurtosis in the options. Meanwhile, in the heart of the distribution, the downside moves from 7% to 11%. Not as dramatic and attributable to market skew.

When we shift probabilities in the tails of distribution vs the meat the impact on the payoffs is significant.

Repeating this insight in a different way may help your understanding. Consider tossing a pair of dice. Imagine playing a game that pays the fair odds for a roll (i.e. craps).


Now let’s chip the dice to change the probability of how they land.

  • In scenario 1, add 1% to the “7” and shave .5% from each tail.
  • In scenario 2, add 1% to the “7” and shave .5% from the meat, the “6” and “8”

By shaving from the tails we take a fair game and turn it into a negative 30% expected value per toss. This is far worse than almost any casino game you might play. By changing the tail probabilities the effect on the game is magnified because the odds are multiplied across an inversely proportional payoff!

Step 2: How should tail sensitivity affect allocations?

By now, the danger of poorly estimating should be a bit more clear. How do we use this when making allocation decisions? After all, most of the time whether they are 1% or 2% events, huge moves are usually not in play. But we must care because when these events hit the impact is huge.

Tail outcomes should dictate constraints based on what you can tolerate. I’ll work through a conservative framework so you can see the impact of naive tail probabilities versus market-implied tail probabilities. The exact answers don’t matter but I’m hopefully offering a way to make tail-thinking relevant to your allocation decisions.

Reasoning through sizing decisions

Suppose things are going well and you are able to save $50,000 per year after paying expenses. You decide that losing $50,000 in the stock market is the largest loss you can accept, reasoning that it’s a year’s worth of savings and that you could make up the lost sum next year. If you impose a restraint like that, well, the most you can allocate to stocks is $50,000. That’s too conservative especially if you have accumulated several hundred thousand dollars in savings.

So you must relax your tolerance. You decide you are willing to accept a $50,000 loss 5% of the time or 1 in 20 years. Roughly a generation. If we use the naive model’s output that we lose 23% of our investment with 5% likelihood then the maximum we can allocate to stocks is $50,000/.23 = $217,000.

The naive model says we can allocate $217k to stocks and satisfy our tolerance of losing $50k with 5% probability. But if the market’s fat-tails are implied more accurately by the option skew, then our max allocation can only be $128k ($50,000/.39).

If we constrain our allocation by our sensitivity to extreme losses, the max allocation is extremely sensitive to tail probabilities. In this example, we simply varied the tail probability between a naive model using a mean and variance to a market-implied model which adjusted for skew and kurtosis. The recommended allocation based on our tolerance dropped a whopping 42% from $217k to $128k.

Many will point out that this approach is extremely conservative. Constraining your max loss tolerance to the amount of money you can save in a year seems timid. But the probabilities we used here did understate the risk. Again these were not peak-to-trough drawdown probabilities but the narrower chance of incurring losses on your start of year allocation. If we are thinking about the true experience of investing and how you actually feel it, you probably want to consider the higher drawdown probabilities which are out of scope for a piece like this. I know many financial advisors read this letter, I’m curious how allocation models reason through risk tolerance.

Current examples to consider in context of small probabilities

1) Bernie

There are market watchers who believe that electing Bernie Sanders would send us back to living in caves. Democrats are trading for about 40% to win the election. Bernie is trading at about 45% to win the nomination, implying an 18% chance to win the election. Market watchers who fear a Bernie presidency are either totally overstating his alleged market impact or the market is already discounting his odds. If the latter is true and the market is efficient, math dictates that it should shoot much higher in the event he loses.

At 18%, Bernie is no longer in the tail of the distribution. So you could argue that as he went from single-digit probability to his current chances, the market strongly re-calibrated either his impact or the sustained rally in the meantime would have been much larger. One of these things must have happened by the necessity of math as odds shifting from a few percents to 18%.

Or there is a third option. The market never really believed that Bernie’s impact would be as deep as his detractors contend.

2) Tesla

We have all seen this stock double in the past month. There has been a lot of talk about far out-of-the-money call options trading on the stock. These are bets on the upside tails of the stock over relatively short time frames. I won’t comment too much on that other than to point out a different tail in the matter. All the credit for this observation goes to a friend who keenly remembered that a year ago the Saudi’s collared their position in TSLA. That means they bought puts and financed by calls sold on the stock. Given the size of the move, the calls they sold are definitely deep in the money. This hedge likely cost them over 3 billion dollars. Billion with a “b”. That’s 6% of there projected government deficit. Their investment in TSLA stock was supposed to be a tail hedge against electric cars destroying demand for oil permanently. In the meantime, they got smoked hedging the hedge. The other tail in this story is going to be that of the official who recommended the hedge. This is a government that nearly executed a 13-year old for protesting. Fair warning to anyone looking to be an execution trader for the kingdom. You are probably short the mother of all puts. Make sure you are getting paid at least as much as a logger.

And one last TSLA note. This keen observation by Professor Bakshi.

Sometimes Keynes’ beauty contest doesn’t just judge beauty. It can create it.