Sam Savage spends a lot of his time thinking about what can go wrong. Lately, he has been working on a new way of working with the statistics of risk simulations.
Savage isn’t a VC or PE expert. He’s a professor at Stanford University’s Engineering School, but his methods for analyzing portfolio risk represents something investors should take note of. He recently wrote “The Flaw of Averages,” (published earlier this year) about how people misunderstand the role of random events and risk in their lives.
PE Week Senior Writer Alexander Haislip recently caught up with Savage to ask five questions.
Q: What mistakes do investors typically make when they look at risk?
There’s a joke about a statistician who tries to ford a river that’s an average of one foot deep but drowns when he steps into a six-foot deep trench. Averages do a poor job accounting for outcomes in investing. That’s particularly true for the venture capital business, where one or two home runs drive enormous returns.
Q: Specifically, what are VCs doing wrong?
A: The venture capitalists I’ve spoken with don’t have a very formal way of looking how the things in their portfolio interact. If you’ve got five crap shoots, and they’re completely independent, it’s one thing. But given a bunch of risks, you can’t just add them up. You have to see how they intertwine.
Q: So what does that mean for a VC portfolio?
Let’s imagine that the big risk driver on your portfolio is the price of semiconductors. The way a typical Monte Carlo simulation would work is that you would go to your ventures and tell them the distribution of risk and they’d calculate various outcomes. But you couldn’t add them together, because the mobile phone company and the pocket calculator company would each be using different probability distributions and you would have a no idea what the real risk would be.
Q: So you’ve created a way to standardize risk calculations for running these simulations?
Yes, I worked with Oracle, SAS Institute and Frontline Systems to develop something called a ‘Distribution String’ or ‘DIST,’ which is a transparent way to represent probabilities. Think of it as a random number string stored in a single Excel cell. The DIST data type could store 10,000 things that could happen to your house and the 10,000 things that could happen with your fire insurance policy. The new data type allows you to sum the probable outcomes of these things and see how they interact.
Q: Using your new data format would be a little like standardizing the discount rate for different investment opportunities, right?
Exactly. Here is a quick analogy. Simulation is to uncertainty as the light bulb is to darkness. It didn’t make night time go away, but it does help you not fall down the steps. Probability is like the electricity that makes simulations work together. A lot of people are doing simulation right, but generating electricity in their backyard. The DIST data type is like the electrical power grid. If a venture capitalist can standardize the probability distribution his portfolio companies can use in simulations, he can get a meaningful grasp of the real risks.