While a correct prediction is certainly better than an incorrect one, in many respects prediction is overrated. Would we admire someone who correctly predicted a coin toss? Then why do we admire people who correctly predict recessions, or elections, or Super Bowl wins?
I’m more impressed by people who correctly predict something for the right reason. Thus while a number of people correctly predicted that there would be a recession in 2008, I don’t recall anyone suggesting that the recession would be caused by tight money. I actually have greater respect for someone who suggested that the performance of the economy in 2008 was uncertain but would depend on the stance of monetary policy. Several well-known people who predicted 2008 correctly did very poorly in subsequent years, an indication that luck may have bene involved.
The 2016 election is another example. I recall several well-known pundits predicting that Trump would win the election. But their reasoning process was flawed. They tended to argue that he would win because his views were surprisingly popular with voters. In fact, the polls were basically correct; Trump was not and never has been popular. Indeed in the 2016 election he won 2.9 million fewer votes that Hillary Clinton. Even worse, Hillary Clinton was herself an extraordinarily unpopular candidate, with very high negatives. Had Trump faced even an average candidate, someone like Joe Biden, the popular vote margin would have been much larger. Against a popular candidate like Obama or Bill Clinton the gap would have been even larger.
Of course Trump did win despite this popular vote discrepancy; my point is that he did not win because he was some sort of “master manipulator” who was able to convince most voters of his merit, rather because his message resonated in geographical areas that are particularly important in our Electoral College system. His enduring unpopularity has been reinforced in subsequent opinion polls, and perhaps indirectly by the recent midterms where Democratic House candidates received 9% more votes than GOP candidates (an unusually large margin.)
I actually have greater respect for pundits who correctly predicted that the electoral college would strongly favor the GOP in 2016, and yet also expected Trump to fall a bit short. (I seem to recall that the “538” website correctly foresaw the electoral college tilt, but still thought Clinton would win.)
What about people who foresaw the recent strength in the economy? How should we evaluate correct predictions on that score? In my view, the strength is partly due to the corporate tax cut, and perhaps to a lesser degree to modest deregulation. On the other hand, the last two times a Republican president came into office with tax cuts and deregulation—1981 and (to a lesser extent) 2001—the economy did quite poorly during the first two years of the administration. In contrast, the last time unemployment was this low (1968-69) was a period of rising taxes and rising regulation. Thus while I thought the GOP policies would slightly boost growth, I did not expect the economy to be this strong. Forecasts should not be based on wishful thinking.
Keynesians have a different explanation for the strength of the economy, pointing to the massive increase in the budget deficit. But once again, fiscal stimulus in 1981 and 2001 was associated with a very weak economy, while fiscal austerity in 1968-69 was associated with an extremely strong economy. So the evidence for the Keynesian view is just as weak as for the supply-side view.
The bottom line is that there is only a very weak correlation between any single policy initiative and the overall strength of the economy. Short-term cyclical moves are mostly driven by monetary policy.
Predictions that are based on false premises don’t impress me at all, just as I wouldn’t be impressed if someone correctly predicted a “heads” coin flip by arguing that, “God favors heads coming up because he likes looking at the bust of George Washington.”
Again, don’t misinterpret this post. I’m not saying a correct prediction is not better than an incorrect prediction–it clearly is. But how much better depends on whether the anticipated causal mechanism also came true. For instance, the failure of the administration to slow the rate of illegal immigration does not mean that a border wall is ineffective, as the border wall has not yet been built. Thus predictions that Trump would fail to slow immigration because border walls don’t work (which is my view) might be seen to have “come true”, but clearly not for the postulated reason. So I deserve no credit for that prediction.
To summarize, I’m not saying that prediction is meaningless, merely overrated.
READER COMMENTS
James
Nov 8 2018 at 5:10pm
You’re describing a classical bias-variance tradeoff.
You’re basically claiming there aren’t enough big event predictions to get a low variance, so you’re going to reduce variance by adding quality of reasoning.
But by also looking at quality of reasoning you introduce bias. Our measure of quality of reasoning tracks predictive ability closely. But it is still our measure, and biased.
Today’s political world is so polarized that I feel it’s hard to be naturally unbiased. Or better yet–to convince others you are unbiased. Because of this partisanship I’m recently more inclined to care about reducing bias than about reducing variance.
nobody.really
Nov 8 2018 at 7:33pm
On probability and prediction:
I deal a standard deck of cards face down into four stacks. I then ask if people would be impressed if the top card of each stack turned out to be an ace. Such as, would they clap for me? Sure.
Then I ask, why?
After some stumbling (and prodding from me), they conclude that it would be impressive because it would be a rare event to occur by chance. Thus, that outcome would suggest that I had employed skill.
Ok, I say, then would you clap if, when I flip over the cards, they turn out to be the 2 of Clubs, 6 of Diamonds, Jack of Hearts and 9 of Spades? No, not so much.
Why not? Isn’t is just as unlikely that I turn up those cards than that I turn up four aces?
After some stumbling (and some mathematical prodding from me), they concede the point.
Indeed, I ask, isn’t ANY combination of four cards just as unlikely as any other combination of cards?
Grudging concession.
So, if you’re going to applaud because I produce an unlikely result, and EVERY result I might produce is equally unlikely, then it shouldn’t matter WHAT cards I turn up, right? Any combination is equally applause-worthy. So let’s just cut to the chase–and I rise and bow, awaiting my applause.
Laughter, but no applause.
Eventually a kid will ask me to flip over the cards. I berate the child: Weren’t you paying attention? Didn’t we just establish that it shouldn’t matter what the cards are, they’re all equally improbable?
The kid still wants to see the cards.
Ok, I say, and adopt a pose of deep concentration. Behold, I reveal to you now—the TWO OF CLUBS! The SIX OF DIAMONDS! The JACK OF HEARTS! The NINE OF SPADES!
None of my predictions proves to be accurate; instead, each of the cards is just an ace. I slam my fist on the table and mutter to myself in frustration, “…stupid trick never WORKS!….”
The kid asks how I was able to do that.
DO WHAT? Make a fool of myself? Long years of practice, believe me…!
But the trick provokes reflection: After all, I expressly promise an outcome that I do NOT deliver, yet people seem impressed with the outcome I DO deliver. They can’t explain why, and they even acknowledge that the act of producing four aces is no less probable than the act of producing any other combination of four specific cards. In short, in the absence of a prediction that I would produce four aces—indeed, in the presence of an explicit promise to produce a different outcome—what makes it noteworthy that I produce four aces?
Robert EV
Nov 9 2018 at 12:30am
When an adolescent I was playing poker with one or two of my half-cousins. We were all dealt royal flushes. My cousin realized he was dealing from a euchre deck. I was still kind of impressed that we were all dealt royal flushes regardless, but assumed that the deck hadn’t been shuffled well (or at all) beforehand.
nobody.really
Nov 9 2018 at 1:09am
That is the PERFECT follow-up post! That’s hilarious–and totally believable.
Bob
Nov 9 2018 at 9:09am
LOL. Years ago I would annoy my friends by selecting 1 2 3 4 in Kino, which you could play in bars. I’d point out that my selection was just as likely as theirs. Kind of ruined the game for them. I’m nicer now.
Benjamin Cole
Nov 8 2018 at 7:33pm
Excellent post.
But worth noting are studies that find that traditional macroeconomic institutions never predict a recession.
This leaves the field of predicting a recession wide-open to quacks.
I wish a worthy academic refuge, perhaps at George Mason, would earnestly devise an early warning system to predict recessions. There is abundant and instant data available on the Internet today.
Perhaps in combination with Scott Sumner’s idea, that the Federal Reserve should be able to make policy decisions much more quickly ( rather than ponderously meeting eight times a year ), then at least monetary policy could become quickly reflexive.
Even then, the solution appears not to be found in QE or interest rates– these mechanisms work too slowly.
Perhaps automatic and triggered elimination of payroll taxes would work. Lost tax revenue would be offset by a money financed fiscal program— the Fed would print money and stick it into the Social Security and Medicare trust fund.
Or, we can wait for the next recession, and then wait for the Federal Reserve to meet, and then wait for the Federal Reserve to institute programs which might slowly reduce the severity of the recession.
You know, like 2008.
Matthias Goergens
Nov 9 2018 at 8:09am
> Perhaps in combination with Scott Sumner’s idea, that the Federal Reserve should be able to make policy decisions much more quickly ( rather than ponderously meeting eight times a year ), then at least monetary policy could become quickly reflexive.
Not needed at all. Rational expectations, level targeting and exploitation of arbitrage opportunities do all the work, so that central banks can take their time.
See eg http://www.themoneyillusion.com/what-is-market-monetarism/ or http://www.themoneyillusion.com/how-many-punches-should-chuck-norris-be-allowed-to-throw/
Matthias Goergens
Nov 9 2018 at 8:13am
No need for the central bank to speed up at all. They just need to be predictable.
Then rational expectations, level targeting and exploitation of arbitrage opportunities will do all the work.
Scott Sumner (and others) have written about this. It’s pretty orthodox economics.
Benjamin Cole
Nov 10 2018 at 4:11am
Matthias:
Yes, in some ways I agree.
But the reality is, someday we will have another recession.
In that case, the Fed will be out of ammo in the first inning, zero-bound wise. Yes, there is such a thing as negative interest rates, but one cannot force banks to lend, or eliminate cash. If my bank says I pay interest on deposits, then fine I keep cash.
That leaves QE. But QE may be a weak tool. No one really knows. The Japan experience is interesting. The St Louis Fed said QE could have been four times larger before influencing inflation rates. Egads.
Sumner deferred to a study that found QE lowered long-term rates by 40 basis points. Wow-wee, that is it? In the Great Recession, the sharpest downturn since the Great Depression, and we want to lower long-term interest rates by 40 basis points? Where is Milquetoast when you need him
Worse, if I suspect the Fed is a weakling, then many others do too. We saw how the Fed handled 2008. So they promise to keep NGDPLT. But we know central bankers care about inflation a lot more than they care about prosperity.
Here is the real deal: The Fed is slow to act, and has weak tools.
Let’s look at money-financed fiscal programs. Read up on Takahashi Korekiyo.
Benjamin Stewart
Nov 10 2018 at 3:12am
Central banks have a monopoly on interest rates. Perhaps a method to have rates set by a market mechanism.
James
Nov 8 2018 at 9:30pm
I would think a utilitarian would be most interested in being able to forecast the consequences of one’s actions as accurately as possible, rather than producing narratives about those predictions.
Scott Sumner
Nov 8 2018 at 9:43pm
Nobody, Nice story.
James, Don’t narratives help to produce reliable forecasts?
Matthias Goergens
Nov 9 2018 at 8:04am
The alternative to revealing your causal model is a long track record of correct predictions. Especially about the same topic.
Thaomas
Nov 10 2018 at 9:01am
The problem of “unconditional” predictions is that the conditions and the model that produce the prediction are opaque, probably opaque to the predictors themselves. Thus they have no policy implications. About the only conditional predictions that were made about the recovery from the 2008 crisis was that Fed policy of QE would produce rapid inflation. It was incorrect, but at lest had the advantage of being falsifiable.
Comments are closed.