Evan Soltas
Jul 6, 2013

Good News, Bad News, and Money

Note: Justin Wolfers writes in, telling me that the optimal k - m is probably slightly positive under uncertainty, as explained by Brainard (1967). That is, the central bank should not close deviations from forecasts completely. Read on for context. 



If you've spent even 30 seconds listening to financial news within the past few years, you're familiar with a line of commentary that seems shallow but might actually teach us a lot about monetary policy.

It's usually phrased like this: Good economic news is actually bad economic news. That's because if payroll employment growth is strong one month, say, that should hurt (not help) equity prices because it implies, on the margin, tighter future monetary policy. Bad economic news, in this view, is in fact good news. Monetary policy will ease to compensate for weak economic indicators.

Or, in the opposite view, it's phrased like this: No, bad economic news is really just bad news, and good news is good. That's because monetary policymakers aren't prepared to fully offset economic data moving forward.

It would help to first note that these are both theories of which effect dominates the other: The marginal bad economic data point or the marginal response of monetary policy to the datapoint. I want to think about this a bit more seriously in this post. But since it's a blog post, not an academic paper, let's make some unusually simple assumptions and see where they take us.

First, your favorite stock market index is in fact a real-time indicator of expected nominal income over an arbitrary future period. (That is to say, in this model, there are no unique effects from monetary policy on financial markets but not on the rest of the economy.) We'll call this N_i in period i.

Next, let's represent the marginal datapoint as δ, which is a real number. (This is a datapoint in the abstract sense -- it's not necessarily payroll employment.) Nominal income is sensitive to the datapoint by a parameter k -- a larger k implies that, before monetary policy, nominal income is more responsible to the marginal datapoint.

Then we also have a parameter m, which stands for the monetary policy response to the datapoint. If we were being fancy, we could have a full monetary-policy reaction m(N,δ), but this will suffice for my purposes here.

A few other assumptions: Time is discrete and counted by i. Parameters k and m are positive real numbers fixed for all periods. There is one new datapoint per period. The new datapoint in period is incorporated immediately into the forecast N_i.

It follows that: N_(i+1) = N_i + (k - m)δ_i. This should be clarifying, as commentaries that argue "bad news is good news" or "good news is bad news" are saying that k - m < 0. Commentaries that argue "bad news is bad news" or "good news is good news" are saying that k - m > 0.

Now let's pivot a bit. What should k - m equal? I say zero. And, allowing N to be the policy variable of choice -- not necessarily nominal income -- you should be answering zero as well.

What does that mean, then? Bad news shouldn't be good news. But nor should it be bad news. It should be irrelevant. The marginal economic datapoint should have zero informational value for a forecast of the policy variable over an interval of time that is sufficiently long and in the future. All I am saying is give commitment a chance -- don't let a burst of good news or bad news alter long-run expectations.

This appears not to be remotely true now. The headline that prompted this post was in today's New York Times: "Jobs Data Is Strong, but Not Too Strong, Easing Fed Fears." Eek. At least in the relevant range of that δ, the argument is that k - m > 0. From experience, equity prices move sharply on the marginal datapoint all the time. And markets for the marginal datapoint appear to price it highly.

Yet I think that for U.S. monetary policy, k - m ≈ 0 if N is price inflation. My observation comes from an examination of the median forecast for quarterly annualized inflation as measured by the Consumer Price Index from 1981 to present. The forecast data comes from the Survey of Professional Forecasters via the Federal Reserve Bank of Philadelphia. The following graph is of the mean absolute deviation from a six-quarter forecast -- which for the forecasters serves as their long-run target -- one to five quarters ahead.

What I find is that the forecasters expect the Fed to erase 63 percent of a deviation in inflation (or for it to just disappear) every quarter. Suppose there's a supply shock, and quarterly annualized inflation deviates 2 percent above target. Next quarter, the deviation should be 0.74 percent, then 0.27 percent, and so on until insignificance. That's a pretty good sign that k - m is indeed in the neighborhood of zero for inflation at six quarters ahead.

Monetary policy has succeeded in making good news and bad news into no news for long-run inflation. That's a historic achievement. But if you believe, as I do, that the optimal policy variable the central bank should stabilize is the path of nominal income, then there is work unfinished.

I see this post as building on earlier notes on why interest rates are rising and why the Fed is having trouble articulating a conditional tapering strategy (see here and here for those). I have to write those posts and do that style of analysis precisely because k - m ≠ 0 for nominal income. It would be a much less nerve-wracking ride if the Fed changed its language from "if the economy heals, then we'll taper, and we'll taper commensurate with the economy" -- which leaves the desired path for the target variable totally vague -- to "here's the desired path of the macroeconomic variable we're targeting, and we'll make our way there regardless of the marginal datapoint, minimizing deviations from the path as quickly as prudent for a central bank."

But, then again, such a monetary policy approach would put most economic journalists out of work. When was the last time you saw a story on price inflation that was not a wire-service summary of new data? What is the ratio of the number of such stories to the ratio of such stories about unemployment, GDP growth, etc.? My answers to those questions are: "2008, maybe?" and "1-to-1000." How about no news at all, then?