A Toronto Data Guy

Paul Graham is right (using AVC’s data)

I don’t like getting into the whole, someone is wrong on the internet thing, but I find the recent discussion that Paul Graham kicked off fascinating: 

[P]ractically all the returns are concentrated in a few big successes. The expected value of a startup is the percentage chance it’s Google. 

He then goes on to say

Some super-angels seem to care about valuations. Several turned down YC-funded startups after Demo Day because their valuations were too high. This was not a problem for the startups; by definition a high valuation means enough investors were willing to accept it. But it was mysterious to me that the super-angels would quibble about valuations. Did they not understand that the big returns come from a few big successes, and that it therefore mattered far more which startups you picked than how much you paid for them?

This has got to piss off some people that invest in startups for a living. Especially coming from a guy that typically gets 6-7% of a company for at or under $30k. I’ll be analysing Fred Wilson’s response below, but first you should read it in its entirety here. (protip: pressing the middle mouse button opens a new tab, I’m still shocked by how few people know this)

On the surface it looks pretty reasonable. He took a 2004 fund, so there should be enough time that has gone by. The first thing that struck me was that he only had two companies go bankrupt during this time. That is outstanding. Fred is clearly an expert investor, his insights are amazing and along with Gabriel Wineberg (who also had a quibble with the Paul Graham post) Fred’s blog is one of the very few I follow outside of what is submitted to Hacker News.

But here it looks like he is wrong. Not only that, he proved Paul Grahams point with his own data. 

The first thing to remember about investing is that you don’t care about 10x returns or 2x returns. I would take doubling my money in a day over 10 folding it over a lifetime, as would any other sensible investor. It is the compounding returns that matter.

So the first thing we need to do with Fred’s graph is convert it into a spreadsheet.

Woah, lots going on there, so let me break down what we have done. I manually counted up the number of companies from the original chart on Fred’s blog post. I’ve assigned a value of 0.5 to the bankrupt companies because I know you at least get tax breaks when you lose money and that in certain circumstances companies can get sold for their on book losses. It might be too high, could only be worth 0.1, but it just helps my argument for it to be 0.5, so I’m going to stick with that. (At least I’m honest.)

Next what we do is convert the 25x and similar returns into their yearly compounded return rates and bucket them into the nearest decile of percentage. Which leads us to…

Maybe not truly bimodal, but not bad for the sample size we are working with. Sure, if I put the value of the bankrupt ones down at 0.1 value they go to -30% per year, but really, Fred isn’t making money on the people that are only worth 1x six years later anyways. So really, his returns are bimodal(ish). 

Really it makes sense that returns are bimodal, especially in software. The cost of the next incremental sale is nearly zero once your product becomes commonplace, so it is natural for a whole host of startups to fail early on (high upfront costs, like developer salaries) and for a few to get into growth stage (highly optimized sales cycles, enough volume for split tests, a recognizable brand, marketplace trust, cheaper capital, CPAs below NPVs, leveraged coder hours, etc) and beyond. 

With a few exceptions, in software you either make it, or you don’t.

I will make one point though, based on some back of the napkin calculations of Google’s Series A investment size, market cap in the year Google went public, and what is openly available of what the founders of Google continued to own (20% each), I’ve estimated Google’s annualized returns from the time they took Series A funding to the time they went IPO to be somewhere between 125% to 200% (that’s annualized(!)). Which is clearly not what Fred is making on his stars. 

This discrepancy is just fine. Obviously Paul Graham doesn’t think that to be a successful Angel you need to get a company that pulls in triple digit yearly gains. His point is that you don’t let the really good ones get away because they are asking twice as much as you were expecting, the Series A venture fund that worked with Google didn’t. Returns are bimodal (or quasi-kinda bimodal). One interesting observation is that Google was notorious for the founders having a large equity stake so late in the funding game. Just one point of data, but maybe good founders know not to give investors an unreasonably large amount of equity.