Tuesday, June 21, 2011

Forecasting models and pre-WW2 recessions

- by New Deal democrat

It seems to be my week to have bones to pick with Berkeley Prof. Brad Delong. On Sunday, quoting a Wall Street Journal article in which Yale Prof. Robert Shiller says he is worried that the US could be lurching towards another recession, Prof. Delong says:
I am not sure why [Prof. Shiller] is alarmed in spite of forecasting models' non-pessimism. So let me tell you why I am alarmed [despite the models' non-pessimism].

One way to think of large-scale forecasting models is that they are essentially vector autoregressions surrounded by a shell of behavioral equations and accounting relationships. The purpose of the equations and relationships is to trigger alarm bells in your mind when you look at the vector autoregression forecast--and then decide whether or not you want to incorporate an add factor. This methodology does, I think, work pretty well for forecasting: it works much better than a poke in the eye with a sharp stick, for example.

But the underlying VAR correlations still largely come from an age of inflation-fighting recessions and rapid bounce-back. Thus that is an input to what the markets are forecasting. And is that input still valid for today?

That is why I am alarmed.
It is puzzling to say the least why Prof. Delong, a student of economic history, is not aware of whether or not forecasting models would operate in pre-WW2 deflationary environments. Most notably, Columbia University Prof. Geoffrey H. Moore spent an entire career gathering data and building indices of leading indicators. When his Institute left Columbia, he founded the Economic Cycle Research Institute (ECRI), which has been very much in the news since their gutsy March 2009 call that the "great recession" would bottom sometime that summer.

Furthermore, specifically in response to criticism by Prof. Krugman, ECRI wrote:
Please note that, contrary to what most bloggers assume, ECRI’s leading indexes are not based on back-fitting of data to “garden-variety” postwar recessions. Rather, they go back over a century and, with no recourse to data fitting, can be shown to correctly anticipate the “jungle-variety” depressions, panics and crises of the early 20th century, as well as the recoveries from those extraordinary events.
(emphasis mine)

)While ECRI's index is proprietary, I have to think that much of Prof. Moore's work, including papers on downturns that preceded the post-WW2 inflationary era, are accessible to Prof. Delong. Thus the claim that at least some forecasting indices do not cover eras like the Great Depression appears simply to be wrong.

Furthermore, the index of Leading Economic Indicators also almost certainly would have accurately forecast the continuing slide of the great contraction of 1929-32, a point I originally made two years ago:
[D]uring the entire period of the Great Contraction of 1929-32, the ONLY time that all three monetary/financial elements of the LEI would have been positive, would have been in September 1931.

Money supply and the yield curve would both have been positive from October 1930-October 1931. Based on how the LEI is presently calculated, those two elements would have been evaluated at about a combined +0.4 each month. The declining stock market would have subtracted 0.1 or 0.2 each of the months (aside from September 1931). Housing would have subtracted another 0.1 or so. So long as the other six indicators, which we have stipulated would have been negative throughout, would have subtracted at least 0.3 each month, then the LEI would have been relentlessly negative throughout the Great Depression until -- appropriately -- close to the end.
---
building permits were presumably also negative throughout the 1929-32 period, except possibly at the very end.
....
What about the stock market? We know what a graph of that, measured by the DJIA looks like, courtesy of dshort. It is safe to say that, with a few exceptions, the three month measure of the stock index would be down except for brief periods in about December 1929, spring 1930, and September and December 1931.

So the bottom line is, if the LEI would be positive, with few exceptions it is going to have to be because of blowout yield curve and monetary conditions.

Well, here is real M1 from Milton Friedman's classic, "Monetary History of the United States." ... [D]uring the Great Contraction of 1929-32, real monetary supply was negative
In fact, at the beginning of 2009 I went beyond that, and wrote a five part series on Economic Indicators during the Roaring Twenties and Great Depression and concluded that the primary difference for indicators between deflationary and inflationary recessions was that:
at no point did either the monetary or the yield curve indicator give a false signal. When either one was triggered, a recession followed.

As I have previously indicated, all the deflationary recessions followed a pattern. The CPI declined from the beginning of the recession and its YoY rate of decline bottomed immediately before the recession's end. M1 followed a similar pattern, sometimes coincidentally, sometimes leading slightly. Like a roller coaster reaching the bottom of a decline, the indicator that the bottom is at hand is that the coaster continues to decline, but declines at an ever decreasing rate! In all 6 of the deflationary recessions during our study period of 1920-50, once M1 and CPI both declined at a decreasing rate, the recession was about to end.
Indeed, if one entirely backs out the yield curve from current readings of the the LEI, the LEI still only show a stall, not a downturn.

Prof. Delong has written extensively in the past several years about the failures of the economic profession. If he believes that post-WW2 business cycle indicators are inadequate, then why hasn't he devoted time and effort to working with the prewar data? After all, Milton Friedman already wrote a Nobel-prize winning book about exactly that data ("A Monetary History of the United States"), and it isn't as if other financial data such as bond rates, housing starts, industrial production, and bond rates are unavailable.

For my part, it is largely bacause of the extensive work I put into dissecting pre-War deflationary era indicators that I believe the economy will be resilient once energy prices stay reasonably moderate, real money supply is positive, and the yield curve is strongly positive. Only strident, deflationary, contractionary economic policy coming from Washington is sufficient to cause me concern that the economy could tip into any prolonged deflationary bust.