Scott Sumner  

Which bus would you take?

I, Needle Nose Pliers... Conspiracy and Condemnation...

Over at TheMoneyIllusion I did a post discussing the issue of whether central banks should be thought of as "controlling" interest rates. Here's one excerpt:

There are powerful cognitive illusions here. The day-to-day Fed "control" over rates creates the illusion of much more control than actually exists. Monetary policy is much more than a series of short run decisions on where to set the fed funds target. By analogy, a bus driver going through the Alps has very good short term "control" over the direction of the bus. He can nudge the bus left or right by turning the steering wheel. But over longer stretches of time the direction of the bus is "controlled" by a combination of the direction of the road (i.e. economy) combined with the bus driver's strong desire than he and his passengers don't hurl over the edge to a terrifying death (i.e. hyperinflation/hyperdeflation).
Let's take that analogy a bit further. You are about to take a bus from Zurich to Milan, right over the Alps. You have three buses to choose from:

1. Bus A is a self-driving machine, fitted with a rear-mounted camera and the latest automatic steering mechanism, designed by noted Swiss engineer Johan Taylor. When the camera sees that the bus has deviated too far to the right of the road, it automatically steers the bus to the left, and vice versa.

2. Bus B is driven by Johanna Yellen, widely regarded as one of Switzerland's best bus drivers.

3. Bus C is a complicated human/machine hybrid. It has forward looking cameras, that feed road images into a large building, in real time. About 10,000 bus drivers sit at the controls of a simulator, and steer the bus as they think is appropriate. The average of all of their steering decisions is fed back to the bus in real time, in order to adjust the steering mechanism. To motivate good steering decisions, the 10,000 bus drivers are rewarded according to whether their individual steering decisions would have led, ex post, to a smoother and safer drive than that produced by the consensus.

Which bus would you take?

Screen Shot 2017-08-02 at 6.19.07 PM.png

Comments and Sharing

COMMENTS (21 to date)
anomdebus writes:

First, use Bus C to gather information, then feed it into a machine learning program. Then gradually transfer from a Bus C scenario to a Bus A scenario by giving the AI more votes relative to the human votes. This process should be reversible, should bugs appear in the AI.

Fwiw, when reading the Bus A scenario, I imagined that people who want to take advantage of arbitrage opportunity may not want too fine or immediate control given to the driver.

I like the Alpine road analogy. It could be extended so that the terrain is also a factor, with the road placement subject to the amount of information known about the terrain.

tpeach writes:

The people in that bus in the picture look very calm and relaxed considering the bus is teetering on the edge of a cliff..... a bit like the members of the FOMC in september 2008....

Jerry Brown writes:

I would take the bus that has the driver inside it even while realizing there is the potential for driver mistakes. The point is that if anything negative, like a bus crash, were to occur then at least the operator will likely suffer the same consequences as me. This may be irrational on my part but I am a person after all.

I hate flying- it makes me nervous. But knowing that there are two pilots on the plane with me helps. If they ever completely automate jet airplanes, I doubt I would ever get on one.

ZC writes:

@Jerry "But knowing that there are two pilots on the plane with me helps."

Unless those pilots are suicidal...then your trust is simply misplaced. But hey, if your irrationality provides you some solace, good for you.

G writes:

If it is a bus: B, hopefully in a few years C, and later on A. If it's a central bank: C, with some caveats.

There is two issues with the analogy: sensors and long term goal. A bus driver is going to collect a lot more information than a simple camera: weather conditions, issues with the bus, environment, etc. A bus driver also has a clear long term goal with some extra incentives: a bonus for fuel economy, tips for smoother driving.

In a central bank situation, you will get a better collection of information using solution C, everybody can look at the main data, but everybody will take into account extra information. The issue with solution C is going to determine what the long term goals are. Should the focus be on global or local issues? Do we want to plan for 1 year, 2, 5, 20 or 50 years? What elements are constituting a good economy? What should we take into account first? Equality? Inflation? Poverty? Health?

RohanV writes:

I assume that to you, Bus C is the free market, and thus the one you prefer.

To me, however, Bus C is a good description of "Twitch Plays...", like Twitch Plays Pokemon, Stockstream, etc. And there's no way in hell I'd trust Twitch chat with my bus.

David S writes:

I think the key here is to add the statement "on the 1000th ride" to each. Otherwise B wins hands down.

On A, until the system is debugged it will be pretty dangerous to ride that bus. On C, the added complexity has the same problems - ie, what happens when the radio fails.

After the bugs are worked out, A will win. It will be cheaper than B or C, and will work "good enough".

rabidwombat writes:

For anyone choosing Bus A, I'd remind you that it is fitted with a rear-mounted camera. There is no mention of a forward-mounted camera - it has to make decisions based only on the road behind it, not the road ahead. In the mountains.

I assume this is a heavy-handed differentiation between John Taylor, Janet Yellen, and Mr. Sumner's free-market-y suggestion. As such, it strikes me as a bit ludicrous.

Thaomas writes:

I'll take Bus with the driver having access to the data from bus C and having to explain why the data of C was not followed if it were not.

Less radically, Bus B whose driver has acknowledged that in the past she had been constrained never to even touch the center of her lane and therefore often had to drive on the shoulder, but that that constraint has now been lifted and she will try to keep the bus in the center of the lane, steering toward the center if the bus drifts in either direction.

Actually a better analogy would be a pilot who has a "dual mandate" both to avoid flying into canyon walls on either side AND to avoid flying into the ground and therefore recognizes that ailerons alone are not enough, that sometime the stick must be used, too.

Don Geddis writes:

@rabidwombat: for all you dislike the post, at least you understood it. Most of the previous commenters seemed to miss the critical details of Bus A.

James writes:

Bus C is a very flattering an analogy for NGDP targeting with a futures market. In the real world, we don't know where the "road" (the ideal path or growth rate for NGDP) even is. A realistic take on is like bus C, except Yellen decides which path is the right one and then the crowd of drivers steers by consensus to get to that path. If the distributed steering system works as intended, the bus goes where Yellen says the road is, even if the actual road is somewhere else. If Yellen is mistaken about the right path for the bus to take, the passengers fall to their deaths and Yellen makes a speech on TV blaming someone other than herself.

Maybe I am being unfair here but I have never seen Scott address the knowledge problem inherent in NGDP targeting. How do we make sure that Yellen picks the right NGDP target?

Will LS writes:

How do you build bus C?

Thaomas writes:

Why cannot monetary policy be driven by the same model that is used to determine the optimal NGDP target?

Bob Murphy writes:

I'm being dead serious: Anyone who answered "C" to Scott's question is having his or her hand forced by prior commitment to NGDP targeting. There's no way in the world you would get on that kind of bus if it were driving through the Alps. You would first want several years of tests on flat county roads.

And I'm not just quibbling with the analogy. For the exact same reason, you should be very wary of NGDPLT proposals.

Dan King writes:

I don't think the analogy works. The road over the Alps is clearly demarcated, making it pretty obvious when a mistake happens. The economic road isn't demarcated at all. There is no map and the passengers can't even agree on where the destination is.

DougT writes:

The analogy actually make a pretty good case for B, without meaning to. Why do you have a driver -- or a pilot -- instead of just depending on autopilot? For emergencies and unexpected developments. Humans may be prone to errors, but they can also innovate and come up with creative solutions to unexpected problems.

All models and algorithms are implicitly short volatility. That goes for machine learning, too. When volatility spikes and we find ourselves outside the system's normal boundaries, I can't always wait for the system to re-calibrate.

Scott Sumner writes:

Jerry, The bad drivers also suffer if Bus C goes over a cliff.

James, You said:

"Maybe I am being unfair here but I have never seen Scott address the knowledge problem inherent in NGDP targeting."

I address it right here. In an uncertain world you rely on the wisdom of crowds, i.e. the market, not experts.

Will, I don't know, it's a hypothetical example.

Bob, i've written papers on how the proposal can be tested, and gradually implement to reduce risk of error.

andrew weintraub writes:

All three buses are driving on a road the curves of which are given and unchangeable. Therefore, the road twists and turns can't adjust to the bus, regardless of whether it's A, B, or C.

I remember Milton Friedman arguing for his rule saying, among other things, that the real economy would adjust to a constant growth of the money supply.

James writes:


You only address the forecasting problem here, but that was not the knowledge problem I raised. Did you neglect to read the question immediately after the sentence you quoted. I asked, "How do we make sure that Yellen picks the right NGDP target?"

I understand (and share!) your enthusiasm for the use of market prices as a better forecast than some macro model. But I think when it comes to NGDP targeting, you are making a mistake you would spot in other contexts.

For example: Let's create a futures contract to forecast the nominal value of output in the auto industry and have Janet Yellen set a target level of output each quarter. The central bank will buy car loan notes until the futures market predicts that total car sales will equal the target.

I'm sure you can think of a dozen good objections to this idea, even while recognizing that the futures market is a good way of pooling information. Guess what: Those same objections apply to having Janet Yellen set a target value for total output.

Jerry Brown writes:

Yes ZC, I am irrational at times. But it really doesn't seem all that irrational to be nervous when you are 20,000 feet in the air and dependent on a complicated machine to bring you down safely. That the operators of the machine are in the same situation as I am does give me comfort.

Michael writes:

agree with Bob's first paragraph, disagree with the second.

If a good bus driver was a good analogy to Janet Yellen, I'd go for her.

The analogy is unfortunate because it misses the important fact that the best Fed president is more analogous to a blindfolded driver using his hearing to pick his way, and in the best case erring towards driving too slow.

I'd say Plato's cave is a better analogy to our situation (surprise!)


Return to top