[hkmurakami]

Be careful to examine what's actually in the fund as your underlying. This isn't like buying the S&P500 where the sector exposure is well defined. One "startup index fund" can perform very disparately from another "startup index fund".

edit: If there's data that shows the performance of a basket of random 100 company samples being equivalent to any number of other random 100 company samples, then I am happy to admit that I am mistaken in this regard.

[swanson]

Yep -- I understand what I'm buying (well, not buying yet...). Something like this would be a great for a Taleb-esque 90/10 barbell I think -- 90% in target-date vanguard funds and 10% in startup investments by proxy.

[kriro]

Didn't Taleb advocate cash holdings (90%) and high risk investing (10%). I always though that 90% cash was a pretty whacky idea and personally just stick to portfolio theory.

I think your plan (substitution index funds for cash) is already a lot better than what Taleb suggested.

[mapleoin]

Taleb considers the stock-market at large to be extremely risky. He advocates the exact opposite of keeping 90% of your assets in the stock market.

[swanson]

Sorry, I wasn't clear. Taleb-esque in the sense of 90/10 (safe/risky) split, not his specific recommendation. Thanks for clarifying :)

[ulkram]

central limit theorum?

[spacehome]

The Central Limit Theorem needs the prerequisite that your distribution has a finite variance. It's not at all clear that this holds for the return of startup investments. In more practical terms, when the variance is very high or there are outsized impacts from small portions of your population, it can take arbitrarily many samples before your average starts to converge. So it's entirely possible that sampling 100 companies isn't enough.

[pyromine]

Statistically the central limit theorem is assuming sampling from the same population, there is no guarantee that two startup indexes are actually sampling from the same population

[ptero]

Lots of conditions that may not be satisfied. Independence is a very hard one. results may be well correlated even if they look like they should be independent