Well there's a rose in the fisted glove
And eagle flies with the dove
And if you can't be with the one you love honey
Love the one you're with
(Stephen Stills 1970)
As I noted in the previous post, I’m working with the Governor’s Woods Foundation to produce a comprehensive data set on private debt for its Debt-Economics project (see www.debt-economics.org). As part of that, last week I realized that the Flow of Funds data on financial sector debt wasn’t what I had hoped it was, and I had to revise down my estimates of the US private debt to GDP ratio from 1952 till today. That revision implied that today’s “Peak Debt” level was lower than that of the Great Depression.
Today’s post revises the pre-Flow of Funds data, using two datasets in the United States Census publication Historical Statistics of the United States from Colonial Times to 1970. One data set is a prototype of the Flow of Funds itself, recording “Net Public and Private Debt” (see Figure 1), while the other is a record of bank assets and liabilities (see Figure 2). The former commences in 1916, while the latter goes back to 1834.
Figure 1: Census Net Debt Table p. 988
Figure 2: Census Bank Assets & Liabilities Table, p. 1019
Ideally, these pre-Flow of Funds records would be consistent with both each other and the later records of the Federal Reserve Flow of Funds. And they are not. But they’re all we have to shed any light at all on the pre-1952 role of credit in the US economy.
It’s not the fault of statisticians that the historical record is so poor; as usual, economists are to blame. Statisticians only collect the economic data that economists tell them are significant, and bank and private debt are ignored in the non-monetary approach to macroeconomics that has dominated the profession for centuries. It’s therefore a miracle that any data is kept at all on debt levels: records of bank assets and liabilities probably owe more to the existence of bank regulators than to economists.
So what do you do when you want a data series that doesn’t exist? You pretty much do what Stephen Stills suggested — make do with what you’ve got. And some statistical plastic surgery is needed to make the best of the absence of consistent data over time.
The first step was to produce a composite measure of nominal GDP over time, since that is needed to compare data from one epoch to another. Fortunately that was easy, because GDP data has been collected in a more or less consistent manner since the late 1700s.
Figure 3: Nominal GDP since 1790
At the point where the series overlap in the 1920s, there’s only the tiniest difference between them:
Figure 4: GDP measures at the overlap in the 1920s
Household debt data is only slightly trickier: the Census measure and the Flow of Funds are almost exactly the same at the end of the Census data set in 1970, but they differ by a small amount in earlier years with the Census measure always being higher (see Figure 5).
Figure 5: Census and Flow of Funds measures of household debt
The easy way to join these two series would be to abut the Flow of Funds data on to the end of the Census data. But the Flow of Funds is today’s definitive series, and that’s the one we have to use for its entirety. So what I did instead was rescale the Census data so that the 1952 value precisely equalled that at the beginning of the Flow of Funds data that year. That of course lowered my estimate of household debt from 1919 (when the Census consumer debt series X-409 began) to 1950. It also meant not using the actual numbers we do have for that period, but it’s necessary to be able to compare the Roaring Twenties and the Great Depression to today: this synthetic series is the best estimate we have of what the Federal Reserve would have declared household debt to be, if they’d started recording the Flow of Funds three decades earlier. That yields what you see in Figure 5, which involves estimating household debt at about 10 percent of GDP less than the Census figures — and less than I had used in previous analysis too.
Fortunately, though this changes the magnitude of previous estimates of the impact of changing household debt on aggregate demand, it doesn’t change the direction of the change in debt at different points in time — and that’s the major issue in working out how debt affects demand.
If only things were so easy for the business debt data: Figure 6 shows just how much these various estimates differ from each other. The saving graces, as with the household debt data, are that the differences are consistent, and the changes in the data are generally in the same direction (though the correlations are low, at around 0.4 to 0.5). Part of the problem here could be recording periods too — the Census data is only annual while the Flow of Funds is quarterly (the correlations improve slightly when the data it lagged one year).
Of the various series, the two that are joined together are X-400, the Long Term Corporate Debt series for the Census, and the Flow of Funds series for Nonfinancial Business Debt. But this requires a major scaling down of the Census-recorded level of business debt in the period from 1916 till 1952.
Figure 6: Numerous conflicting estimates of business debt
That makes a huge difference in the estimate of the level of business debt during the Great Depression: from a deflation-charged peak of 140 per cent to a “mere” 85 per cent — which is only slightly higher than the peak level reached in 2009 in our current crisis.
Figure 7: Comparing previous and new estimates for business debt 1916-1952
So much for the easy stuff. Now we step back into the Wild West — literally and figuratively — of data on debt during the 1800s.
The Census Department does have one series on this — X-582, its record of bank loans from 1834 until 1970. But it’s hugely different to both the Flow of Funds and the Census data itself for “Net Indebtedness” of the household and business sectors.
If I had my druthers, I’d rather the bank loan data be the assets measure of the gross indebtedness of the household and business sectors (including non-bank financial institutions), and the household and business debt data be the liability measure of the same thing. But again I have to “love the one I’m with”, and that’s not what I would wish, obviously, from Figure 8.
Figure 8: Rescaled Loan and Original Bank Debt Data
However, as you can see, the turning points in the two data sets are very similar for the range in which they overlap (and the correlation coefficient is a respectable 0.76). But the levels are very different (though the smallest difference applies for the earliest data). The implication is that, had the Flow of Funds been developed when Wyatt Earp was a not even a kid, it would have produced a higher estimate for private debt than the bank loans data imply.
The end result is that Figure 9 is the best estimate we have of the long run private debt data for the United States. Curiously, it restores the ranking that I had believed applied to our current crisis when I was erroneously including some bank debt (as opposed to loans) in my data: our current crisis involves the highest level of private debt in US history.
Figure 9: Estimated Private Debt for the USA
Much as I would like it to be, this probably won’t be the last word on revising the data on US private debt. We still have to compare our data to other sources (especially Moritz Schularick’s 15 country, 150 year study), and I’d like to tease non-bank debt out of the dog’s breakfast of data on financial sector debt recorded by the Flow of Funds. And if there are other groups out there that also have estimates on pre-Flow of Funds debt, we’d love to hear from you.
Ultimately, what I’d really like is for statisticians to do this work for us. But that will require convincing the economics profession that the level and change in private debt is an economic issue. Given, as usual, Bernanke’s complete failure to consider private debt in his “100 Years of the Fed” speech, that day is still a long way off.
Figure 10: "… pure redistributions should have no significant macroeconomic effects” (Bernanke 2000, p. 24)