Comment on Del Negro, Giannoni & Schorfheide (2014)
I would like to try to critique Del Negro, Giannoni & Schorfheide (2014) “Inflation in the Great Recession and New Keynesian Models”
h/t Brad DeLong.
Before the jump I have to say 3 things
1) I have just skimmed the paper. I didn’t work through the equations.
2) I am very hostile to the whole discorso (roughly literature or research program)
3) I am more favorably impressed than I would like to be.
I will now attempt a super brief abstract (I am terrible at this — click the link).
Del Negro et al 2014 builds on an earlier paper which attempts to explain the collapse in output in2008-9 as the result of an increase in risk premia due to an exogenous increase in the variance of entrepreneurs’ ability. The model fits GDP. The new contribution is an effort to explain the small decline of inflation and absence of deflation. The average rate since 2008 of inflation is close to their forecast and closer to their fitted value given Fed policy (including foreward guidance).
In the paper, there isn’t anything special about zero inflation — cutting prices is just like raising prices. Del Negro et al contest the claim that some special nominal rigidity at zero change is needed to fit the data.
Now I will try to critique.
Rational price setting with sticky prices (Calvo pricing) implies a very important effect of expected inflation quite far in the future on current inflation. The agents in the model believe that the Fed will find a way to keep inflation near 2% and this causes inflation to be near 2%.
In case anyone is reading, Calvo pricing is the assumption that firms can only change prices when their Calvo alarm clock rings. Opportunities to change prices arrive at a constant rate, so the change that the firm will have no choice but to charge it’s current price t units of time in the future declines exponentially with t. Also the interval since the firm last had a chance to change it’s price is exponentially distributed.
I have two (hostile) questions. First the model fits the mean but does it fit the variance of the first difference ? Calvo pricing with a low arrival probability of opportunities to change prices makes inflation very smooth. Del Negro et al write that they do not fit or attempt to fit high frequency fluctuations in inflation. However, their model has strong implications for such fluctuations. If taken seriously, the varince of first differences is a summary statistic or stylized fact which must be fit. Seriously the puzzle is why did annualized quarterly core PCE inflation bob up and down in the range 0.2% to 2.2% but never fall below zero. This just does not look like a variable with huge intertia. It looks like a ball bouncing on something near zero.
Second, how well does the model fit the 70s and early 80s. the parameters (including the Calvo arrival frequency). I think the model will forecast very poorly starting at 1973, 1975 or 1981.
One key feature of the model (which I’m sure makes it possible to reconcile the data from the 70s and 80s with a low Calvo frequency) is that economic agents’ beliefs about the Fed’s target inflation rate are modelled as time varying (an AR1). Since the target inflation rate is very important to firms whick look long into the future, it is possible to fit all sorts of time series of inflation ex post by choosing the time series of perceived target inflation.
In effect the story of the 70s and 80s is one of the bold Volcker regime shift which caused a dramatic change in inflation by causing a dramatic change in expected future inflation. Here there are implications for variables other than inflation — in the 70s and 80s explicit forecasts of inflation from surveys. In this milenium those forecasts and TIPS spreads. These implications are not tested. In fact, the Livingston panel of “expert forecasters” systematically over estimated future inflation when Volcker was Fed Chairman. The standard story does not fit any data except for those on achieved inflation. This is also true if it is told with math and by a computer.
Also since the target inflation rate changed due to an exogenous shock, the forecast will be of little change in inflation. That looks right for the past 5 years. It looks terrible for the 70s and 80s.
Risk premia are central to Del Negro et als (and DeLong’s) explanation of the great depression. The paper does not confront the risk premia forecast by the model for 2008-2014 with the time series of actual risk premia. I think the model achieves the forecast of a slow recovery by modelling risk premia so they stay very high even though actual risk premia returned to normal. If I understand correctly, Del Negro et al. see if they can fit the data using an ex post model of Fed policy. I don’t see a reference to plugging in observed risk premia.
I have an even crankier complaint about the financial frictions. They are modelled as the effect on risk premia of exogenous and otherwise unobserved variation in the variance in skill across entrepreneurs. Like the Calvo alarm clocks this isn’t meant literally. Since this variable appears only as a shifter in the risk premium, I think the micro foundations add nothing and subtract nothing.
The model does have a clear implication. The skill varies across entrepreneurs not workers. Financial frictions should affect firms not households. The spread between say Baa bonds and Treasuries should vary but the spread between the mortgage interest rate and Treasuries shouldn’t (nor should underwriting standards for mortgages ever vary). If one divides fixed investment into residential and non residential — the model implies persistently low non residiential fixed investment and normal recovery of residential fixed investment — the opposite of what has happened.
This argument will certainly be considered unfair. It is obvious that by “entrepreneurs” the followers of Bernanke Gertler and others don’t mean entrepreneurs. The model is designed to fit total investment and abstracts from minor details like the existence of a housing sector. The only implication of the micro founded model is that risk premia can vary for unexplained reasons and risk premia affect investment. My objection is that, since in practice all deviations between micro founded models and an ad hoc aggregate models are bugs not features, what possible use could there ever be in micro founding models.
I have just made a very broad claim. I can’t think of an exception.