Scott Sumner  

There's no success like failure

Against Argumentative Definiti... Henderson on Economic Inequali...

I've worked in several different types of organizations, and I've observed a growing tendency for managers to look for measurable metrics to assess productivity. Fortunately, I've generally been in a position where I didn't have to spend a lot of time on those reports, but others are not so fortunate. (That's not to say these management techniques are not useful or effective, I suspect at least some of them are valuable. Management is not my forte, to put it mildly.)

Consider someone working on monetary policy at a research institute. What might constitute a successful performance? One could envision a set of intermediate objectives, such as publishing research on NGDP targeting, or engaging with important people in academia, the media and government. Maybe writing an op ed for the NYT or WSJ. Then there are the ultimate goals, to influence Fed policy with the force of your brilliant ideas.

In the vast majority of cases, the intermediate objectives are positively correlated with the ultimate goal. The correlation may be really weak (my NYT editorial would be unlikely to influence the Fed), but the correlation will be at least slightly positive. There is one case, however, where I sort of wonder whether the correlation is negative. I'd like to get your thoughts on this as well.

One of my recent projects is to encourage the formation of the Hypermind NGDP market. We now have NGDP prediction markets for 2017:Q1 to 2018:Q1 NGDP growth, as well as another market for 2018:Q1 to 2019:Q1 NGDP growth. These two markets are currently showing an expectation of 4.7% and 4.4% NGDP growth. Has this project been successful? That turns out to be a quite interesting question.

The intermediate objectives of this project might include a high volume of trading, as well as productive academic research on the correlation between fluctuations in NGDP futures prices and other key macro variables. The ultimate goal of the project might be to get the Fed to stabilize NGDP growth at roughly 4%. And here's the weird part---these two goals might well be negatively correlated, indeed strongly negatively correlated.

First a bit of background. I started blogging in early 2009, mostly because NGDP had recently been extremely unstable, and I believed that this instability reflected very poor monetary policy and had huge negative effects on the economy. For whatever reason, the subsequent 9 years have seen (along with the 1990s) the most stable NGDP growth in all of American history. I'd like to believe those two events were connected in some way, but I doubt even our supremely self-confident President would be so bold as to make that sort of grandiose claim. No, more likely I just happened to start blogging at a point in time where monetary policy became much more stable.

The problem with stable monetary policy is that it makes monetary blogs much less interesting, especially monetary blogs whose sole raison d'ĂȘtre is to talk about the problems associated with unstable NGDP. Thus my MoneyIllusion blog has seen a steady deterioration in quality and readership. So how should I feel about that?

In a similar fashion, an NGDP futures market is not likely to attract much attention from traders, as long as it keeps chugging along, year after year, with predictions of roughly 4% NGDP growth. Nor will it attract much interest from academic researchers, or financial reporters. So how should I feel about that?

If my work for Mercatus and future Fed policy are truly 100% uncorrelated, then the answer to the preceding puzzle is clear---I should prefer the success of the Hypermind NGDP prediction market experiment. After, all the success or failure of that experiment will have no causal impact on NGDP or monetary policy more broadly. (Yes, it will still be correlated, but no causal impact will occur.)

But now suppose there is a tiny chance that the Hypermind NGDP futures market will attract enough attention to the need for stable NGDP growth expectations that it nudges monetary policy slightly toward a more stable NGDP growth path. Say a one in a million chance. Surprisingly, that changes everything. This market is so inexpensive, and the benefits of stable NGDP growth are so vast (think trillions of dollars), that even a tiny chance of it influencing policy is enough to make us root for its failure. That is, we want to root for the market continuing to show almost no volatility, which means there will be little interest in the project.

Digression for nerds: This is actually quite similar to Newcomb's Problem, as brilliantly explained by Eliezer Yudkowsky in this post. I'm not sure that Yudkowsky is correct, indeed I suspect that the way the problem is generally set up there is some missing information, so I don't know if there actually is a "correct" answer. In Newcomb's Problem there is an explicit statement of the problem that implies that choosing one box over the other has no causal implication. But since the person predicting which box you'll pick has amazing accuracy, we are led to suspect that the ground rules are somehow incomplete, that there is some sort of "spooky action at a distance" even though we are given ground rules that suggest no way for that to occur. So (Eliezer claims) we take the box with $1,000,000, not $1,001,000, as a way of insuring against the risk that we've overlooked something important. Or at least that's how I interpret his argument. It's well worth reading the post.

Back to my post: So let's say the Hypermind futures market has a tiny probability of influencing Fed policy. Then I might root for its failure. Or perhaps it's just a sort of "watched pot never boils" form of superstition on my part. I have a low opinion of my ability to undertake any successful project in life, so I sort of feel I'll fail at this too. I'll keep asking people run an expensive project year after year, which over and over again fails to produce any sort of interesting results.

But unlike all my other failed projects in life, this one has been cleverly set up to have a silver lining. Let's say the project does fail, again and again, for the rest of my life. What does that mean for my ultimate objective, a Fed policy that successfully keeps NGDP growing at roughly 4%/year?

Unless I'm missing something, it can only mean one thing:

Screen Shot 2018-02-19 at 6.09.58 PM.png
PS. Here's Eliezer Yudkowsky, channeling Al Davis:

First, foremost, fundamentally, above all else:

Rational agents should WIN.

Don't mistake me, and think that I'm talking about the Hollywood Rationality stereotype that rationalists should be selfish or shortsighted. If your utility function has a term in it for others, then win their happiness. If your utility function has a term in it for a million years hence, then win the eon.

But at any rate, WIN. Don't lose reasonably, WIN.

Comments and Sharing

COMMENTS (3 to date)
Rajat writes:

I'm not sure what you mean by 'failure' of the experiment. Surely what you want is for the Hypermind market to predict a rapid slowdown in NGDP, for the Fed to ignore it and 'do nothing' and for a recession to follow. You can then spend the rest of your life shopping the need for the Fed to target market forecasts of NGDP. That outcome might require a failure on the part of the Fed, but not of the experiment.

Scott Sumner writes:

Rajat, I'd rather have the Fed keep targeting NGDP at 4%, and the experiment fail due to a lack of interest.

Yes, a successful NGDP market would be better for me personally, but the achievement of a successful monetary policy is much better for society. I'd have to be a pretty amoral sociopath to root for a successful NGDP prediction market experiment!

Peter McCluskey writes:

I'll suggest measuring Hypermind success by some sort of citation count. Preferably one that heavily weights citations from Fed researchers. I'm unsure whether mentions in the news media should count as positive or negative - I'd certainly want the ratio of news media mentions to Fed mentions to be low.

>choosing one box over the other has no causal implication

No, naive heuristics about causality suggest no causal implication, but the evidence about omega suggests naive causal models are wrong.


Return to top