Experiment HN: Crowdsourcing Tech Stock Predictions

It seems clear to me that the pricing mechanism is not readily understood by anyone, hence the ability for quantitative models to make money and yet still not be pure money printing machines (with the exception of HFT which does not claim to understand the general pricing mechanism, only short time span deviations from prior prices regardless of whether the prior prices are "wrong" or "right"). Further, the quantitative models I'm aware of price things on a relative basis : "given that we derive these general market parameters, things should be priced like f(x)", as opposed to "this company is worth X dollars". That fundamental analysts have jobs while still differing widely in their opinions suggests to me that the way to price things on an absolute basis is not well understood.

The underlying aggregation function may be simple, but we can't know that since its not transparent. My vote there goes towards not simple though.

Its most certainly democratic in some ways, though its unclear how votes are allocated. One per individual? One per trade? One per dollar? One per unit of influence?

I think the idea of crowdsourcing from a subset of the market could provide some edge, but think this is the wrong subset.

I'll argue that equity markets are not good platforms for crowds to judge the fair market cap of companies (or the NPV of cashflows). Most participants have an incentive to maximize profits or preserve capital subject to some risk appetite, and very few are actually interested in pricing companies; instead participants are mostly interested in what they think the others players are doing.

If we were to crowdsource what movies HN users enjoyed, that could work. However, if we said that you would be paid for your vote only if you were in the same group as the majority, then suddenly you aren't voting honestly... you are trying to figure out how other people might vote, and our final result is the aggregation of a lot of game theory (plus a few naive honest players who get picked off by the rest of us).

Therefore, equity prices are not a democratic vote. Prices are not a weighted sum of your vote times your conviction (nor are they your vote * conviction * wealth). They can break down to a battle of speculators trying to outwit each other and exploit any flaws in the market or others' thought processes.

D. E. Shaw Research recently set a record simulating a single protein molecule in surrounding water for about 1 millisecond using first principles from computational chemistry, and I suppose that means they have a long way to go before fully modeling protein folding because the aggregation function of all those forces are complex. However, I'd bet it would be much harder to aggregate all the market participants and all their incentives from first principles to step 1 millisecond forward in live trading time.

I'm not saying you can't make profits. Obviously people have for a long time found relationships here and there that can predict future price movements. I'm saying a formal aggregation model that would let us fully compute and Turing-decide future prices, even if we had all the relevant inputs, seems complex.

Edit: I'll post aggregate values to my blog in a few days' time. zacharyburt.com. I'll cc the link to @zburt on Twitter.