Sunday, May 01, 2011

Stolen From Brad's comment thread.

Finally I get a response. Brad DeLong tells me what would convince him that QE2 was a flop

If market expectations of future inflation had not risen but had declined since QEII was mooted last August, I would conclude that QEII has been ineffective. They have not.


Well that was brief and to the point. He also had a graph showing the difference between the 10 year treasury yield and the 10 year Treasury Inflation Protected Security (TIPS) yield. This can be roughly interpreted as expected inflation and it has increased as anticipated following QE2.

I reply

Robert Waldmann said...
Hmm on my hard disk I conceded that one. But I think the bottom line reports real interest rates. They have not changed.

If expected inflation had fallen a little and 10 year nominal interest rates had fallen a lot, would you conclude that QE2 had failed ? Or are you playing heads I win, tails lets flip the other coin ? Do you think that nominal interest rates don't matter at all, that is that only expected inflation matters not real interest rates ? Would I have been delighted that my debt was inflating in 1982 when inflation was around 10% and the 10 year Treasury rate was 15% ?

I don't see any possible benefit from a simultaneous increase in nominal interest rates and expected inflation. How does that affect anything real ?

I now guess that there might be a point that the nominal interest rates we care about are those paid by corporations or mortgagers and that it is possible that purchases of treasuries reduced the price of treasuries relative to other assets.

Here is another one. Let's take an IS-LM model with nominal interest rates on real GDP (just to be as barbaric as possible). QE is a shift down of something like an LM curve. The explanation for increasing nominal interest rates is that the IS curve shifted up because of increased expected inflation and increased expected GDP growth. The model would give a reduced real interest rate, but let's forget that. What about now ? The explanation from uh someone who I won't name (not Brad) is that sure growth was bad in Q1 but it would have been worse without QE2. That is, the IS curve shifted down. So the evidence is a raw correlation (with total flexibility about the timing ascribed to QE2). Then the actual correlations of QE2 with interest rates and real GDP are both explained appealing to omitted variable bias. I say no. One shouldn't run simple regressions and then dismiss correlation coefficients. But even if I bent over backwards to be generous to (that person) I would have to bend over forwards too, since he (or she) assumes that the IS curve simultaneously shifted up and shifted down.


In the comments thread SvN raises the spirits of Neyman and Pierson from the vasty deep

SvN said...
I'm troubled by the notion of "proving something ineffective."[*] May Bayes forgive me, I think in terms of rejecting a null hypothesis of no effect. Is Waldman asking us to believe that QEII has had a significant negative effect on the economy? or that its effects have been significantly below what was expected based on our models? or simply that there is no significant evidence of positive effects? Power and size and very different animals, as are Type I and II errors. Which are we worrying about here?



Good questions SvN. I phrased my post carefully as a question. I didn't ask anyone to believe anything. I asked people (very especially including Brad also by Gmail) to tell me what would convince them.



As you note, proof that an effect isn't positive can consist only of statistically significant evidence that it is negative. This is the spirit of Brad's answer too. He said that if expected inflation had decreased he would be convinced that QE2 had failed. He didn't say that if it had remained the same he would be convinced (for one thing it doesn't stay exactly the same for long since it is the difference between two asset prices and one of the assets is traded a lot).

I am convinced that QE2 failed. As you argue, I must look for data which seem to show it had the opposite of the desired effect and not just data which show no significant effect. I am convinced by two time series -- the 7 year constant maturity Trasury nominal interest rate (I look at the daily series from FRED) and the ugh quarterly (Ugh) series of investment. The QE2 works hypothesis has three implications

1) 7 year nominal interest rates should be reduced (the policy is massive purchases of 7 year treasury notes so the most obvious and direct effect should be an increase in the price of those notes)

2) expected inflation should increase (I think this should happen indirectly via the forecast effect of interest rates on GDP and the effect of GDP on inflation via the Phillips curve which is stubbornly still in the data long after economists decided that believing in it was like believing that the Sun orbits the Earth). The two effects should both cause market real interest rates to fall, that is the price of TIPS (Treasury Inflation Protected Securities) to rise.

3) lower medium term real interest rates should cause increased investment and, especially, increased investment in housing. So the hypothesis leads one to predict that the growth of investment (and especially in housing) should be above the trend rate of growth.

So what is the score.
1) 7 year nominal interest rates have increased since QE2 was first mentioned by Bernanke (August 26) and since the final QE2 program was announced (mid November) and during the period of actual QE2 purchases. The price of the asset of which the Fed is buying $600,000,000,000 worth went down. I think this is strong evidence that the Fed can't drive that price up (wrong sign not just insignificant).

2) a Tie. I look at the 7 year constant maturity TIPS rate (7 years because the purchase was of regular un indexed 7 year notes). It's close to what it was on August 26th. Since mid November it has increased. My effort to psychoanalise the time series leads me to speculate that some investors thought that QE2 would work (obviously many economists did) but when the actual purchases started people stopped being willing to put their money where their mouth was as it became clear enough that QE2 was a flop that no one would bet on its success (while many people are perfectly willing to declare it a triumph since words are cheap).

3) extreme disappointment. It is a bit early yet. Still QE enthusiasts were declaring victory long ago. The latest numbers are early for an effect of a shift in monetary policy on GDP and components -- the very conventional view is that the effect reaches its peak in about 2 quarters so really the current quarter 2011q2 not the one (2011q1) whose numbers have just been added up). But investment was dramatically below what was forecast in August 2010 and November 2010.

So on the three reasonable statistics the score is 2 to 0. Two dramatic changes with the wrong sign and one very close to no change.

I don't claim that the null that QE2 worked at all is rejected at the 5% level. The data are very few or very noisy. That's why I ask people and not a computer. The answer may be the following "I am only convinced when a null is rejected at the x% level -- therefore I have never been convinced by any claim in empirical macroeconomics. The data are few, disturbances are heteroskedastic and serially correlated, the parametic classes of models such that we can calculate distrubutions under the maintained hypothesis that the data generating process (DGP) is a member of that class clearly don't include the DGP. So no data which have been or can be actually collected will ever convince me of any claim in macroeconomics."

Fine. That's damn close to my personal view. But that wasn't Brad's answer. He said, in effect, yes of two asset prices one moved the wrong way and one moved hardly at all. But the only thing that matters is the difference between those two asset prices, which moved the right way. My follow up question is why do we care about the difference ? Don't we care only about the asset price which didn't move ? I can promise you, the logic of the pro QE2 argument applies to real interest rates and not to the difference between real and nominal interest rates. Also I claim that there is almost exactly nothing there. Certainly no statistically significant evidence against the claim that QE2 worked at all.

I have another post on the topic. In that post, I note the shreds of evidence that QE2 had the opposite of the intended effect (as you note the only way the evidence can prove that it didn't have the intended effect). I explain that I don't find that evidence convincing at all. So, in answer to your question, I certainly don't ask anyone to conclude that QE2 failed.

Introspecting, I can go further. In my heart I am sure that QE2 had effects of the anticipated sign -- lower nominal interest rates that would happen otherwise, higher expected inflation, and higher investment especially in housing. I certainly don't claim that this inner certainty is in any way based on evidence. I am also convinced that the effects were much smaller than hoped (and anticipated really). So I definitely don't think that a true coefficient is negative or exactly 0. I think it is significantly less than was believed. But I can't give you a number for what I think was believed, or a point estimate of the coefficient. I haven't even explained what coefficient I have in mind (hmmm I guess the interest rate elasticity of demand for 7 year notes). I sure can't tell you at what level I believe the one tailed hypothesis "QE2 works at least as much as anticipated" is rejected. I'm pretty sure the level is above 5% (that is the null which I dislike, criticize, and insult is not rejected at standard signficance levels). Macro data are crap, and it's early yet.




Please don't read this stupid footnote

* I never wrote the words which SvN put in quotation marks. SvN typed "'proving something ineffective'" I did *not* type "proving something ineffective." SvN misquoted and libeled me. SvN's whole point is that someone who understands classical statistics wouldn't have written "proving something ineffective" as I didn't. I firmly assert that, when one discusses something written by someone else, there are only two possible uses of quotation marks. One is to quote that person word for word with ellipses in the place of any deleted passage (as SvN did not do). The other is to lie or show reckless disregard for the truth (as SvN did).

No comments: