DISQUS

robdiana Alex, I know I get benefits to this guest posting as well. Louis has a much larger reader base than I do, so I get a little more exposure than normal. Hopefully, everybody wins in the end.

Alex Hammer Not that you need it but you have my permission to identify that that is me (to which this post refers). Suffice to say that there are two sides (or more) to every story, and I'm not going to get into that ditch.

Best wishes.

Preston Sumner And we've seen how Wall Street has used those predictive models to great effectiveness! I do like your point about baseball GMs - it suggests an inefficient market resistant to change. Since Billy Bean's success there may not be much new advantage to be gained. Read Taleb's "Fooled by Randomness" and "Black Swan" for some great insight into the pitballs of expert predictions.

iamverytall A couple comments below. Overall, I'm not exactly in disagreement - VC's could certainly (and almost certainly do) do some modeling to help them figure out their potential ROI. However, I think that there are some considerations that should be taken into account.

1. Predictive modeling is useful only up to a certain point. As I'm sure you've heard many, many times by now, the current credit crisis (as well as other crises, e.g. the LTCM implosion) can largely be attributed to blind, unwarranted faith in the power of predictive models. You could potentially argue that this was just because the models weren't powerful enough - perhaps they failed to take into account what Soros calls the "reflexivity" of market behavior. But things like that are essentially black swans - you don't know that they exist until your investment has gone belly up.

2. What most major Wall Street firms are spending their money on is risk management, i.e. figuring out how much money they could possibly lose on a given investment, as opposed to what they are likely to gain. The models that tend to work best are the ones where there is a significant/huge amount of historical data to test against, e.g. corporate default data for corporate bonds. As the amount of data decreases, the reliability of the model begins to break down. This is why Moody's had no real business rating CDOs and other recent structured products - there simply was no way that they had enough data to accurately model the risk on those instruments.

There are a couple of points here. First, because risk models look at established structures, they should (theoretically, at least) have a fairly solid amount of quantifiable information about the underlying asset. Credit analysts can look at audited financial statements, etc. for use in their analysis of a company's credit-worthiness. It's not immediately clear that a VC can do the same, especially for early-stage companies that are unlikely to have detailed financials.

Moreover, it strikes me that a lot of the "secret sauce" in venture capital investing is not in the quantifiable aspects of the investment, but rather in more qualitative views on macro-trends, management teams, the potential for disruptive change in established business sectors, etc. While it is not impossible to ascribe a value to these elements and place them in a model of some sort, the degree of accuracy that you'd expect to see is so low that you might as well just toss the whole thing out and pick your final number out of a hat. Garbage in, garbage out.

3. It's not necessarily clear that VC firms have the resources to do this sort of thing on their own. Major Wall Street firms are much, much larger than major venture capital firms.

I'd say more, but this is getting really long for a comment. Again, I'm not saying it's impossible. I'm just saying that it would be hard to build a model that you could rely on to accurately predict the ROI on a particular investment. You can (and probably should) put a handful of assumptions into a model that might give you a view on the likelihood of your investment's success. But in my view, these models should always confirm hypotheses, rather than drive them.

Taylor Davidson The evaluative mechanisms used to evaluate VC investments are based on an understanding of the present and future. A startup typically creates or requires some fundamental change to succeed (change in technology, economics, marketplace, changing consumer behavior, changing management, etc.).

Predicting the fundamental change necessary to extract the required value from the noise, in that scenario, simply requires too much "art" to make the "science" like that meaningful.

It's not hard to figure out the main things VCs look for in investments: tying those variables into a model requires too much judgment and assumption to make the results valuable.

Alex Hammer Theoretically, such algorithms can be devised via reverse-engineering. You look at the ROI results of your or others investments, and then you work to identify and quantify the contributing elements to that result. You keep tweaking the algorithm until it best fits (predicts) the result. Then you move on to other results to see whether the algorithm just created also (best) predicts this other result, whether it needs to be further modified, or whether this new result requires its own separate (but perhaps related) alogorithm