Tuesday, April 23, 2013

The Ongoing Dereliction of Duty

Last year I made the case that the Fed's failure to keep nominal income growth expectations stable was a dereliction of duty:
[We] have long made the case that a nominal GDP (NGDP) level target would firmly anchor the expected growth path of nominal income.  Doing so, in turn, would stabilize current nominal spending since households and firms are forward looking in their decision making.  For example, holding wealth constant, households generally will put off purchasing a new car or renovating their homes if they expect their nominal incomes to fall and vice versa.  This is why Scott Sumner likes to say monetary policy works with long and variable leads. This understanding implies, therefore, that the reason for nominal spending remaining below is its pre-crisis trend is that the Fed has failed to restore expected nominal income to its pre-crisis path. This failure amounts to a passive tightening of  monetary policy.
Since then, the Fed has improved its management of expectations by introducing the conditional asset purchasing program of QE3. While this program is progress, it is still far from adequate. This can be easily seen by looking at data from a question on the University of Michigan/Thompson Reuters Survey of Consumers where households are asked how much their dollar (i.e. nominal) family incomes are expected to change over the next 12 months. The figure below shows the average response for this question up through March, 2013:


The fall of household dollar income expectations and its failure to fully recover is stunning. It suggests that the now lower expected future income growth is depressing current household spending, a point forcefully made by Mariacristina De Nardi, Eric French, and David Benson of the Chicago Fed. Digging into the data, they find that expected nominal income growth deteriorates across all age groups, educational levels, and income levels over the past few years. This is not some sectoral-specific development, it is a systemic nominal problem. They also find that the collapse in expected dollar income growth explains much of the decline in aggregate consumption since the crisis erupted.

But there is more. The figure also indicates that real debt burdens are higher than many households expected prior to the crisis. Look at the dashed line. It shows the average expected dollar income growth rate over the 'Great Moderation' period was 5.3%. Now imagine it is early-to-mid 2000s and you are taking out a 30-year mortgage and determining how much debt you handle. An important factor in this calculation is your expected income growth over the next 30 years. If you were average, then according to this data you would be forecasting about 5% growth rate. But that did not happened. Household dollar incomes declined and are expected to remain low. Nominal debt, however, has not adjusted as quickly leaving higher than expected real debt burdens for households.

This is something that the Fed could correct. QE3 is a step in the right direction, but more needs to be done with this program to raise expected nominal income growth. One way to do this is to make the size of the asset purchases conditional. That is, instead of conducting fixed $85 billion purchases every month until the economic targets are hit, vary the size of the purchases depending on the progress of the recovery. For example, if inflation and unemployment are not moving fast enough to their target, then increase the dollar size of the of asset purchase and vice versa. For if $85 billion is not enough for the nominal economy to gain traction, then it must be the case that money demand is rising enough to offset the benefits of the $85 billion injection. If this conditionality were added and widely understood, QE3 would better manage expectations and pack a larger punch. No more dereliction of duty.

Update I: Per Nick Rowe's request I have added the following two figures.  The first one shows expected household dollar income growth plotted along side NGDP growth over the past year. The former does seem lead the latter.  


The second figure shows the mean and the median expected household dollar income growth. Interestingly, the gap between the two series is relatively stable until the crisis, after which it narrows.


Update II: The first two figures above use a three quarter center moving average to smooth the series. The last figure--the one directly above showing the mean and median--shows the raw, unsmoothed series.

25 comments:

  1. Very important post David. I didn't know that data existed.

    The message speaks fairly clearly. The one discordant note is that expected nominal income growth starts to fall a little earlier than I would have expected -- earlier than a simple "the decline in expected NGDP is what caused the decline in actual NGDP" hypothesis would suggest. How did actual NGDP manage to keep growing for as long as it did?

    A suggestion: superimpose a plot of actual NGDP growth on top. Let's see how well they match. Expected NGDP must be a large part of the story. But it can't be the whole story (not that surprising, really).

    It's also a little puzzling that there are only slight signs of earlier recessions (except 82, of course).

    ReplyDelete
    Replies
    1. Nick, I have added the graph above in an update. I also added recession bars so that they can be seen easier.

      Delete
  2. Nick
    Actual nominal income fell below trend almost as soon as Bernanke took over in January 2006 and the distance to trend just kept rising...before tumbling.

    ReplyDelete
    Replies
    1. I see it the same way, but i still find it a bit curious that 2005 was "the year"

      Delete
  3. "How did actual NGDP manage to keep growing for as long as it did?"

    Falling labour share of income might be a good hypothesis, Nick. Even today, actual NGDP growth is outpacing expected dollar income growth.

    ReplyDelete
    Replies
    1. Falling labor share of income had a lot to do with it. As well, in the late 90's, there was a drop in unit labor costs. and then a surge in unit labor costs before the recession of 2001. Then unit labor costs fell after the recession. Since the recession of 2001, labor share has steadily declined.

      At the same time, effective demand had abruptly stagnated and even fell during the early 2000's. There was still excess effective demand, though, encouraging businesses to grow. Once that excess effective demand began to run-out around 2006, the crisis loomed. The increase in aggregate profit rates slowed down, even though profits were good. The expectation of an ever increasing economy began to disappear. Utilization rates for capital and labor stopped improving for 3 years before the crisis.

      Also, inflation dropped after the 2001 recession leading to a drop in the Fed funds rate. Lowering inflation and a low Fed funds rate together meant lowering unit labor costs which translate into lower income expectations. Nominal GDP kept growing in at atmosphere of growth and surging investment.

      Delete
  4. I applaud the Fed for trying to do something in the face of an economically illiterate Congress. Unfortunately the above post also nicely illustrates the limited influence that the Fed can have on GDP. As David Beckworth rightly points out, what needs to be stabilised is household “nominal spending”, and that spending will fall when households have difficulty dealing with their debts.

    QE, i.e. stuffing cash into the pockets of the rich or creditors in exchange for their Treasuries just doesn’t touch the above problem. What would touch the problem is fiscal policy, i.e. feeding cash into household pockets.

    ReplyDelete
  5. I am very skeptical that central banks can always manage expectations in the way you suggest. Nick Rowe has convinced me that targeting NGDP would be better than targeting the interest rate, but not that targeting NGDP would have prevented the current U.S. recession. If the central bank hits the zero lower bound, it doesn't matter which you target. You may well not be able to hit either target however much you try. The bank of Canada may have been able to hit its inflation target through this recession, but that was only because commodity prices were being driven up by Chinese demand. I doubt if it could have hit an NGDP growth target of 5%. It just had no interest rate room to do so. And if it couldn't do it, no-one is going to believe it could next time.

    Quantitative easing is only able to bring other interest rates down to zero. That could help a bit, but there's still a floor. If you hit that too, you can't do any more.

    The only available solution that lies within the rules that have become established is fiscal stimulus. Let the government borrow the money the central bank is printing and either cut taxes or spend it on things that put money into the bank accounts of real people. That's how to stimulate demand.

    ReplyDelete
  6. It's the animal spirits that Keynes described almost a century ago! Say it loud and say it proud: "We're all Keynesians now!"

    ReplyDelete
  7. Granger causality tests suggest that Consumer Survey median income expectations does not cause NGDP but NGDP does cause Consumer Survey median income expectations.

    I wanted to repeat the tests with Consumer Survey average income expectations but the data file I downloaded didn't have that information. It had proportion of survey respondents by ranges of response (in addition to the median response).

    Precisely how do I find the Consumer Survey average income expectation?

    ReplyDelete
    Replies
    1. This comment has been removed by the author.

      Delete
    2. I computed the moving average of the expected income for every quarter. Neither GDP nor expected income's MA are stationary and they do not cointegrate either. If you take first order diferences both series become stationary and no granger causality exists either way at any significance level. From a VAR(k,k) on the stationary series, various results appear the higher the k. In general, the VAR adjusted R^2 of regression in which D(gdp) is the dependent variable is much larger than the adj. R^2 of the D(exp. income) being the dependent variable.
      Moreover, from a VAR(2,2): accumulated generalized impulse responses of D(exp-inc) to a D(gdp) permanent innovation although significant fades away within a year, while accumulated GIR of D(gdp) to a permanent D(exp-inc) innovation has permanent effects.
      On the other hand, if you take the non-stationary series, GDP appears to granger cause expected income (quarterly moving average).
      Any thoughts?

      Delete
    3. Something more: over the last 15-20 years the impact of D(exp-inc) on D(gdp) is far more larger and more statisticaly significant while the reverse causation is weaker and more statisticaly unsafe (VAR(2,2)).

      Delete
    4. Mark and Alex: this sounds interesting. But we might want to distinguish between:

      1. Which caused which over the whole time period?

      2. Which caused which in one particular episode (like the recent recession)? (Harder to test of course, except by the eyeball method?)

      The theory that changes is ENGDP *always* cause changes in NGDP wouldn't make sense, unless expectations were totally irrational, or totally self-fulfilling in a world of multiple equilibria?

      Delete
    5. Alex,
      I did my analysis with median rate of change in income expectations and the annual rate of change in quarterly NGDP using Eviews. I find that income expectations is not stationary but that its first difference is stationary (as is NGDP).

      I set up a two equation VAR in the levels of the data including an intercept for each equation. The various information criteria all suggested a maximum lag length of 2 for each variable. However an LM test for serial correlation suggested it was necessary to increase the lag length to 5. An AR roots graph suggested it was dynamically stable and Johansen's Trace Test and Max. Eigenvalue Test both indicate cointegration between the two series at the 1% significance level.

      Then I restimated the levels VAR with one extra lag of each variable in each equation. But rather than declare the lag interval for the 2 endogenous variables to be from 1 to 6 I left the interval at 1 to 5 and declared the extra lag of each variable to be an exogenous variable.

      When I did the Granger causality test I failed to reject the null that income expectations does not cause NGDP but I rejected the null that NGDP does not cause income expectations at the 1% significance level.

      I also repeated the tests with year on year change in NGDP and the results were essentially identical.

      I stopped my analysis at that point because the results seemed fairly conclusive.

      Delete
  8. Hey Mark, good to hear from you again. First, the folks at the Chicago Fed have a new paper where they test the predictive power of the expected nominal income changes: http://www.chicagofed.org/digital_assets/publications/economic_perspectives/2013/1Q2013_part2_french_kelley_qi.pdf

    Second, to get the average you have to download the raw survey data (look for downloading cross sectional option)and then collapse the raw data into an average. I did it in Stata using the 'collapse' command. You can also email me if you want the data directly.

    ReplyDelete
    Replies
    1. Yes, I noticed that James Pethokoukis did a post on that paper last week after I read your post. (I was looking for a more detailed description of the average income expectations series.) Incidentally, I noticed that James mentions this post in his most recent post.

      I was at the raw survey data screen but had trouble downloading it. I think it will be simpler if I just send you an email. Thanks.

      Delete
    2. I completed the Granger causality tests on the data you sent me and what I find is:

      1) "with 95,-95"
      A) quarterly NGDP Granger causes average expected income at the 1% significance level
      B) average expected income Granger causes quarterly NGDP at the 10% significance level
      C) year on year NGDP Granger causes average expected income at the 5% significance level
      D) average expected income Granger causes quarterly NGDP at the 5% significance level

      2) "without 95,-95"
      A) quarterly NGDP Granger causes average expected income at the 1% significance level
      B) average expected income does not Granger cause quarterly NGDP
      C) year on year NGDP Granger causes average expected income at the 1% significance level
      D) average expected income does not Granger cause quarterly NGDP

      The second graph in your post is of year on year NGDP and, apparently, of the "with 95,-95" expected average income series, so the Granger tests suggest causality runs in both directions at the 5% significance level.

      Delete
  9. Mr. Beckworth,

    You wrote that the Federal Reserve has committed "dereliction of duty" by failing to target a "nominal GDP (NGDP) level...[that] would firmly anchor the expected growth path of nominal income".

    First of all, it is not the Federal Reserve's duty to increase nominal income much less its expected growth path. The goals of the Federal Reserve are "... to promote 'maximum' sustainable output and employment and to promote 'stable' prices.. There is nothing in there about increasing anyone's income or expectations of income.

    Second, targeting a NGDP (the McCallum Rule) will no produce different policy results than targeting inflation (the Taylor Rule). In the current economic situation, both rules would produce the monetary policy, low interest rates, maximum liquidity, and the purchase of government bonds, notes, and bills.

    The Federal Reserve is not tasked with increasing income expectations and has no power to influence them.

    David de los Ángeles

    ReplyDelete
  10. This is something that the Fed could correct. QE3 is a step in the right direction, but more needs to be done with this program to raise expected nominal income growth. One way to do this is to make the size of the asset purchases conditional. That is, instead of conducting fixed $85 billion purchases every month until the economic targets are hit, vary the size of the purchases depending on the progress of the recovery. --Beckworth.

    Excellent suggestion. Right now, no one believes in the Fed, it has no credibility, thanks also to various FOMC members who race around in sweat-drenched hysterics about inflation and make florid public proclamations.

    If the Fed laid out a program of rising QE, and said it would stick to it come hell or high water...maybe it would get more traction.

    ReplyDelete
  11. I'm starting to think QE3 is a mistake, not because monetary policy is too loose but because QE3 with a 2% inflation target is like flooring the gas while also standing on the brake. Expectations uber alles!

    ReplyDelete
  12. The impotence of QE is nicely illustrated by the fact that despite the unprecedented increase of central bank or “exogenous” money over the last four years, endogenous money in the US is the same as it was four years ago. See chart here:

    http://azizonomics.com/2013/04/25/even-after-all-the-qe-the-money-supply-is-still-shrunken/

    ReplyDelete
  13. The figure also indicates that real debt burdens are higher than many households expected prior to the crisis.

    Net Worth stopped rising in 2005-6 as house prices peaked. Construction started shedding jobs. (including millions of illegals which are NOT captured in the graphs. Real employment was already dropping in 200, but this was mostly illegals for a year)

    After 20 years of finance and manufacturing becoming finely tuned to produce new house construction, the builders succeeded in outtsripping ability of new buyers to pay. Unlike the quite fast dot.com bubble pop, or the '29 Crash, the house price peak was followed by many months of slow drift down, as people who still had jobs and income only slowly "realized" that their Net Worth was less than they had been expecting. When the house speculator flippers starting selling, the small glut turned into a 10-20-30% price drop, all of which came out of buyer Net Worth.

    Ag Demand is based on "how much money do buyers have (for spending)".
    There was a huge, negative Net Worth shock. Prior expectations, as house investors as well as economic producers, turned out to be wrong. Builders by the thousands (tens? hundreds of thousands?) stopped building, many going belly up.

    This huge, economy wide mal-investment was coupled in a perfect storm with the new phenomenon, first time since WW II, of baby boomers starting to retire. People who retire expect their HouseHold income to drop, since it does. The BB retirement wave was ALWAYS going to include an expected HH income drop. This means the USA will never return to the prior, 45 year post WW II baby boomer "trend", unless there is another baby boom.

    ReplyDelete
  14. Its very informative blog....the description about debt marketing is not so good to me but through the graphs one can easily understand easily.

    ReplyDelete