[huac]

Black-Scholes (and the Merton version) are useful as simple estimates of option prices. Of course it's not going to be accurate (it assumes a lognormal distribution for volatility, JFC) - but the model is public and easy to calculate, providing a stable basis for these kinds of prices.

Edit: people buying/selling options are of course performing their own pricing operations. They don't care how the market price is determined, since they believe their model is the best and gives more accurate prices than anyone else.

Edit 2: furthermore, to properly price startup options, you need to account for their 'random-expiration' nature: we have no idea when Uber will IPO. this makes it difficult to use traditional option pricing models which have fixed maturities (you can take integrals over the model's results at each possible maturity). additionally, choosing a discount rate is hard when realizing that most employees are not well-diversified, unlike the investors/funds that the traditional models are written for

[scurvy]

There's not much difference between an American style public equity option and private equity option. Both can be exercised at any time. Perhaps you're thinking of European style options, that can only be exercised on a fixed date (expiry).

While neither are great, binomial is better for American style options than BS.

[huac]

American options still have fixed expiration dates, and cannot be exercised after that. While that's closer to a startup's option structure, it still doesn't account for the uncertainty in when or if the startup will go public. Additionally, you never exercise dividend-less American options early, and startups generally don't offer dividends.

I guess you could choose an arbitrarily distant expiration date, but my bigger point was that while BS is clearly not going to give perfect results, the results are reasonable enough and transparent enough to justify their use for tax purposes in lieu of actual market prices.