Y Combinator Numbers
I was recently asked in an interview how YC is doing. We're old enough now (6 years) and have enough data (316 startups including this summer's) that we should be able to start to answer that.
What should we measure, though? The easiest thing to measure is how many of the startups we fund raise more money after YC. By that test we're doing fairly well. Even before Ron Conway and Yuri Milner ensured the number would be 100% by offering $150k to every startup we funded, we'd got it up to about 94%.
There were 36 startups in the summer 2010 cycle, the last one before Ron and Yuri started funding everyone. Of those, 33 raised more money after YC, 1 didn't bother because they were already so profitable, and 2 were not able to raise money. So 34/36 or 94.4% either raised money or didn't need to. That number is about as high as I'd want it to be. If it were 100%, I'd worry we were being too conservative in who we funded. 
But while funding is easy to measure, it's not the real test, either for us or for the founders. What matters is how the companies end up doing. Getting funded is not success. It's just something that makes success more likely. Which means if you measure how well the companies do, you're already measuring everything that matters about funding.
How do you measure how well a company is doing? The standard test of that is its value. So the best measure of something like Y Combinator is the average value of the companies it funds.
Ultimately you want to measure the values at "exit," meaning either acquisition or IPO. The valuations of funding rounds before this point are effectively attempts to predict it.
The problem with measuring exits is that they take a long time, and the most successful companies tend to take the longest. Even 6 years is not long enough to generate accurate numbers. We've now had 25 companies acquired, 5 of them for over $10 million, but the total value of the remaining companies dwarfs the total value of those 5. 
So the best we can do at this stage is to use the acquisition prices of the companies that have been acquired and try to estimate the current values of the rest. That's not as hard as it sounds, because like any portfolio of startups ours has a pretty steep power law distribution. If we can produce accurate estimates of the values of the top 10%, we'll have a sufficiently accurate estimate of the total value.
10% of what pool though? I found in practice it was pointless to consider companies we funded less than a year ago. Most of the startups from winter 2011 don't even have valuations yet, because they raised money on convertible notes. So the last batch I considered was summer 2010. We'd funded 208 startups up to that point.
The combined value of the top 21 companies is $4.7 billion. Which I admit surprised me. 
$4.7 billion / 210 = $22.4 million, so the average value of startups we've funded is about $22.4 million.
Of course that could be cut in half tomorrow. Startup valuations are even more volatile than ordinary equities. But at least we now have a measure of how we're doing. And unless the top VC funds as a whole lose money, this number should be a lower bound on actual exit valuations. 
The real lesson here though is how long it takes to measure performance in this business. We're 6 years in, and we could easily be off by 3x in either direction. Startup outcomes are unpredictable, and the outcomes of their investors doubly so, because it's hard to say whether the big successes are repeatable, or if the investors just got lucky. Even 6 years in, all we can say is that the numbers look encouraging so far.
 We're not looking for companies that have a high likelihood of getting funded after YC. We're looking for companies with high variability. E.g. we're ok funding groups for whom the likelihood of failure is high but for whom success, if it happens, will be big.
 When startups are acquired, acquirers usually give the founders incentives to stay on. Sometimes to make the founders feel better these incentives are included in the number the acquirer quotes as the acquisition price. In that situation there are effectively two prices, the quoted price, and a lower price investors see. Obviously we use the lower of the two.
 For 18 of the top 21 I used the postmoney valuation of the most recent funding round. The other 3 had grown significantly since their last round, and to estimate the values of those I used the opinions of independent experts and the valuations of the funding offers they are currently receiving.
 Lower-tier VC funds do as a whole lose money, but so far at least the great majority of VC investments in YC-funded startups have been by the better firms. I'm not sure exactly why, since bad VCs are more numerous than good ones; perhaps the bad VCs assume the companies at an event as public as Demo Day will already have been picked over by the time they get a shot at them.
Historically, top-tier VC funds have had positive returns. But
they could end up losing money if the whole economy tanks.