My Conversation with Hal Varian

Hal of course was in top form, here is the audio and transcript.  Excerpt:

COWEN: Why doesn’t business use more prediction markets? They would seem to make sense, right? Bet on ideas. Aggregate information. We’ve all read Hayek.

VARIAN: Right. And we had a prediction market. I’ll tell you the problem with it. The problem is, the things that we really wanted to get a probability assessment on were things that were so sensitive that we thought we would violate the SEC rules on insider knowledge because, if a small group of people knows about some acquisition or something like that, there is a secret among this small group.

You might like to have a probability assessment of whether that would go through. But then, anybody who looks at the auction is now an insider. So there’s a problem in you have to find things that (a) are of interest to the company but (b) do not reveal financially critical information. That’s not so easy to do.

COWEN: But there are plenty of times when insider trading is either illegal or not enforced. Plenty of countries where it’s been legal, and there we don’t see many prediction markets in companies, if any. So it seems like it ought to have to be some more general explanation, or no?

VARIAN: Well, I’m just referring to our particular case. There was another example at the same time: Ford was running a market, and Ford would have futures markets on the price of gasoline, which was very relevant to them. It was an external price and so on. And it extended beyond the usual futures market.

That’s the other thing. You’re not going to get anywhere if you’re just duplicating a market that already exists. You have to add something to it to make it attractive to insiders.

So we ran a number of cases internally. We found some interesting behavior. There’s an article by Bo Cowgill on our experience with this auction. But ultimately, we ran into this problem that I described. The most valuable predictions would be the most sensitive predictions, and you didn’t want to do that in public.

And:

COWEN: But then you must think we’re not doing enough theory today. Or do you think it’s simply exhausted for a while?

VARIAN: Well, one area of theory that I’ve found very exciting is algorithmic mechanism design. With algorithmic mechanism design, it’s a combination of computer science and economics.

The idea is, you take the economic model, and you bring in computational costs, or show me an algorithm that actually solves that maximization problem. Then on the other side, the computer side, you build incentives into the algorithms. So if multiple people are using, let’s say, some communications protocol, you want them all to have the right incentives to have the efficient use of that protocol.

So that’s a case where it really has very strong real-world applications to doing this — everything from telecommunications to AdWords auctions.

And:

VARIAN: Yeah. I would like to separate the blockchain from just cryptographic protocols in general. There’s a huge demand for various kinds of cryptography.

Blockchain seems to be, by its nature, relatively inefficient. As an economist, I don’t like this proof of work that this is. I don’t like the fact that there’s one version of the blockchain that has to keep being updated. I don’t like the fact that it’s so slow. There are lots of things that you could fix, and I expect to see them fixed in the future, but I would say, crypto in general — big deal. Blockchain — not so much.

And finally:

COWEN: Now, users seem to like them both, but if I just look at the critics, why does it seem to me that Facebook is more hated than Google?

VARIAN: Well, you know, I actually don’t use Facebook. I don’t have any moral objection to it. I just don’t have the time to do it. [laughs] There are other things of this sort that can end up soaking up a substantial amount of time.

I think that one of the reasons — and this is, of course, quite speculative — I think that one of the reasons people are most worried about Facebook is they don’t really understand the limits of what can be done at Facebook. Whereas at Google, I think we’re pretty clear that we’re showing you ads. We’re showing you ads that are targeted to one thing or another, but that’s how the information’s used.

So, you’ve got this specific application in our case. In Facebook’s case, it’s more amorphous, I think.

There is much, much more at the link.

Comments

Why might Facebook be (far) less loved than other tech tyrants?

Pieces like the following may offer explanatory power:

https://www.cnbc.com/2019/06/19/facebook-content-moderators-expose-disturbing-working-conditions.html

FB seems a poorly managed company that fails to monitor its vendors adequately. (Does this mean that FB employs "poor management algorithms"?)

Straussian reading: Companies do lots of secret prediction markets

If Facebook has more and deeper information about its users, why does Google capture a much larger share of digital advertising revenues (their combined share is over 60%)?

With algorithmic mechanism design, it’s a combination of computer science and economics.
------
Strip the economic theory to the basics. Then implement the theory in a computer network, substituting bots for humans. Then, as the final step, let the humans own and control the bots. Bingo, no need for economists, we will compute it ourselves, thank you.

One way to use prediction markets is to leak to the press that management is mulling a decision to acquire another business, or put itself up for sale, or undertake a strategic review of the portfolio of businesses, or whatever. Then see how the market reacts and what kind of sell-side research reports come out, usually with commentary about “our conversations with investors suggest a split would unlock value,” that sort of thing. That is pretty common but not for day-to-day business decisions, small deals, or even pretty sizable capex commitments, and is not done repeatedly across the whole decision space, so it’s a limited tool.

Why would or how can the locution "prediction market" be deemed precisely accurate?

Don't economists "actually" mean and intend "probabilities market"?

My pedestrian reading of contemporary physics suggests that The Future enjoys no baryonic constitution whatsoever: we thus have no predictions to make, we have only better and less informed guesses to make as we extrapolate.

If The Future already exists (more or less as our devotees of Progress prefer to maintain), someone might want to tell the rest of us.

"There are lots of things that you could fix, and I expect to see them fixed in the future, but I would say, crypto in general — big deal. Blockchain — not so much."

This isn't informed thinking. Flip the notion around and make the same statement that Blockchain is real and crypto is not and you've got the same amount of substance.

"The problem is, the things that we really wanted to get a probability assessment on were things that were so sensitive that we thought we would violate the SEC rules on insider knowledge because, if a small group of people knows about some acquisition or something like that, there is a secret among this small group."

This sounds profoundly unlikely. So there's an amazing efficient tool which could revolutionize management and works, but - oh no! *Insider trading*! Gasp! We can't have *that*: the world would end if there was a little more insider trading.

Rather, this sounds exactly like Robin Hanson's explanation: management doesn't want prediction markets because they create a track record which exposes management failings early on in a legible way, exploding management excuses, and so management come up with excuses to not use PMs to protect themselves. You might say that the problem with PMs is that there are secrets that 'a small group' are keeping, and they do indeed have to do with 'insider trading', in a sense, but that PMs threaten to expose those secrets to 'outsiders'...

I doubt your naive answer is correct. I bet Google's C-suites really are concerned with going to jail for insider trading expressed in prediction markets. It's another example IMO of how bad laws (including bad patent laws) are holding back innovation.

Bonus trivia: prediction markets don't always work, for the same reason the stock market has correctly called 9 out of the last 5 recessions, and if the USA escapes a recession this year, it will further validate that popular sentiment is not an accurate predictor of a recession (i.e., expected AD as stated ex ante to an actual recession via consumer sentiment and popular opinion is not that important, further reinforcing there's something to business cycles that are endogenous not exogenous)

Ray nails this one.

C suite (hell VP and even director levels) at Google are ludicrously compensated. The idea of doing anything that could land a person in jail is completely discarded.

I’m sorry Rick, you have no idea what you are talking about.

Because if we look at Google and Facebook executives' decisions over the past decade, we see that they are terrified of insider trading because any of them have been prosecuted for it, and we see that they are truly scrupulous about following the law and do absolutely nothing which might go anywhere near the line, no matter how much profit it profits?

Sorry, but you guys are clueless. How much more blatant self-dealing does an excuse for not using prediction markets have to be? Did all those two-level share structures also get put in place out of the love of securities law?

I misread this as "My Conversation with Van Halen" and now I really want to hear that podcast.

Could you imagine a Tyler interview of Daimond Dave?

Outrageous.

Maybe we'll hear about it later.

On predictions:
Credit ratings indicate the risk of borrowers defaulting.
Bankers when determining the appropriate exposure and risk premiums consider that perceived risk.
Therefore “risky” borrower are not dangerous to bank systems.
Since they seemingly all missed the lecture on conditional probabilities the current regulators, very unfortunately for us all, do not understand this.
https://subprimeregulations.blogspot.com/2018/08/risk-weighted-capital-requirements-for.html

Textbooks are okay in price on Amazon. I can find them cheaper elsewhere, always.

Until it hits free.

And then I can see my textbook at the end of the year for some obscure price...

And then I've made money on a free book.

Students are perhaps the most rational consumers when it comes to finding textbooks online.

He is so eloquent.

"By the way, I know a few billionaires, and there’s a lot of cost to being a billionaire in the sense that you can’t go out in public. Maybe you need bodyguards. Doing a trip here and there is a major undertaking because of the people that have to be informed. It’s much better to be a half a billionaire, I think, than to be a billionaire."

Good advice. Thanks, Hal.

"Why do people trade so much in financial markets?"

I'm curious how many of these trades are due to income inflows... e.g., I get paid every two weeks, I buy a few shares of this or that... it looks like I'm doing a lot of "trading," and indeed a lot of trades are occurring, but conceptually they're really all just one big trade, spaced out over time...

Another reason why one might not do prediction markets is that they may involve controversial matters where some might think a prediction coming out of one might help lead to a bad outcome. This is why Robin Hanson got in trouble once upon a time for doing a prediction market on the probability of an assassination of the king of Jordan, which led him to be denounced publicly by members of Congress and Paul Wolfowitz because it was claimed this might encourage somebody to actually follow through on doing that.

Glad my question about AI driven tacit price collusion made the list. As a consumer I'm not that enthused about his response though.

I think what's driving his skepticism is that any market with an active interest currently has a regulated over-the-counter or exchange-listed contract. Regulators will make sure anything vaguely similar will be fully regulated, removing all crypto-related advantages (eg, permissionless access, immutability). Predicting events with obvious stock price implications would fall under that, and the insider trading angle is just one of several rationalizations for why it wouldn't be allowed without the full vetting one would apply to a Swap Execution Facility. The other stuff--predicting elections, whether the next British baby is a boy--is allowed, but they're insignificant.

Then why aren't prediction markets running on blockchains more popular? It is not like they share Hal's concerns of being regulated.

I was arguing why they are not popular, and thus not of interest to Google. Augur and Gnosis have prediction markets that exclude everything people want to trade to avoid the SEC clamping down on them. No one can trade size their because no corporation can do so without breaking some regulatory rule. So all you have left are individuals trading things like Dai into ETH, or binary options on the price of Ethereum that settle months after expiration. There are many other problems with these services related to costly delays from their 'consensus mechanisms', which allow them to argue they have no control over the remaining offerings, but these are very slow--it takes a month to get paid for a SuperBowl bet--and still buggy (many users on Augur game their system by creating technically invalid markets due to ambiguous wording).

There's no upside for Google entering this game.

I think Google has a lot of good-will just because (1) it makes a lot of great products for consumers (gmail, google drive, android), (2) it makes a lot of great products for programmers (google colab), and (3) it spends a lot of cash on programmers: every year they pay $5000 for several thousand people to work on free software projects (it's called summer of code), at least in the past they paid pizzas to computer science meetings (it was called the pizza embassor).

The only thing Facebook does that is parallel to that is in the area of machine learning. They do publish quite a bit and they develop a tool that is pretty popular called PyTorch (albeit the Google version is much more popular, TensorFlow).

Comments for this post are closed