Sunday, June 7, 2015

Fancy maths and data series are no reason to ignore supply and demand

Came across a 2014 NBER paper Betting The House (pdf) by Òscar Jordà, Moritz H.P. Schularick & Alan M. Taylor. I was wildly unimpressed. I am not quite sure whether I am willing to use the tag line of "numbers make smart people stupid"--as per this wonderful post on the adoption of farming, criticising an attempted cliometric study of said transition from foraging--but still, wildly unimpressed.

Judging monetary policy fail
First of all, the paper associates low interest rates with "loose monetary conditions". For example:
The long-run historical evidence uncovered in this study clearly suggests that central banks have reasons to worry about the side-effects of loose monetary conditions. During the 20th century, real estate lending became the dominant business model of banks. As a result, the effects that low interest rates have on mortgage borrowing, house prices and ultimately financial instability risks have become considerably stronger... these historical insights suggest that the potentially destabilizing byproducts of easy money must be taken seriously (p.3).
Or, later:
Using short-term interest rates as a proxy for monetary conditions ... (p.14).
The paper also refers to "easy low interest rates" (p.35).

The paper is centred around an equation based on the monetary policy trilemma (that monetary policy cannot have a stable exchange rate, free flow of capital and autonomous monetary policy all at the same time) where interest rates "measure" the stance of monetary policy (p.15), leading to the central conclusion (p.37) that (italics in original):
Loose monetary conditions are causal for mortgage and house price booms, and this effect has become much more dramatic since WW2.
Judging the stance of monetary policy from interest rates is deeply problematic. In Milton Friedman's words:
Initially, higher monetary growth would reduce short-term interest rates even further. As the economy revives, however, interest rates would start to rise. That is the standard pattern and explains why it is so misleading to judge monetary policy by interest rates. Low interest rates are generally a sign that money has been tight, as in Japan; high interest rates, that money has been easy.
This is hardly surprising, as nominal interest rates include inflationary expectations, so will be higher if inflationary expectations are higher. During the Great Moderation, inflation and interest rates were low: in what world is low inflation a sign of "loose monetary conditions"? To quote Milton Friedman again:
After the U.S. experience during the Great Depression, and after inflation and rising interest rates in the 1970s and disinflation and falling interest rates in the 1980s, I thought the fallacy of identifying tight money with high interest rates and easy money with low interest rates was dead. Apparently, old fallacies never die.
Apparently, they don't.

Money has two uses--the demand to hold money as an asset and the use of money in transactions. (Or, if you like, the demand to keep money for use in future transactions and its use in current transactions.)  If inflation is low, that means that the use of money in (current) transactions is roughly keeping pace with the growth of goods and services. So, not loose monetary conditions, with (nominal) interest rates to match. (Which, by the way, is the only way the quantity theory of money makes sense--i.e. if the demand to hold money is not included: money held has no effect on the price level because it is not being used to buy things.)

As long as one separates out the demand to hold money as an asset/for use in future transactions, then monetary conditions are a matter of supply and demand (of money in circulation and goods and services on offer). A simple way of thinking of the price of money is that it is what you can buy with it, so 1/NGDP (the inverse of total money spent on goods and services). Thinking of it that way, the price of money moves inversely to the price level (since NGDP = Py, where P is the price level and y is output of goods and services) holding y constant: as the price level goes up, the price of money in goods and services falls (and inflation is when that keeps happening); as the price level goes down, the price of money in goods and services rises (and deflation is when that keeps happening).

Conversely, if the demand to hold money shoots up, as happened in 2008, but the central bank does not adjust monetary policy accordingly (as also happened in 2008) then the effect is a serious (if passive) tightening of monetary policy. Which can lead to things such as the steepest fall in spending on goods and services in the postwar period. All that without any significant shift in interest rates.

Exchange rates are also prices of money. But they are the price of one money in terms of another money.

What is NOT the price of money are interest rates. (The price to borrow something is not the same as the price to buy it.) Interest rates are the price of credit (aka, the price of delayed obligation). Low interest rates (which normally implies low real interest rates) are a sign that credit is cheap, so we can expect, in such circumstances, more purchasing of goods and assets on credit. Such as, for example, housing.

To put it another way, easy credit is not the same as easy money. Though changes in a monetary regime can certainly shift risks. For example, when unification into a common monetary area lowers nominal interest rates in previously inflation-and-exchange-rate-depreciation prone economies that are no longer subject to differentiated money risk (since they are now using a single currency) nor to exchange rate risk (ditto). But that is not really "easy" money; it is a shift in expected risks favourable to more use of credit. One, indeed, based on the expectation that the common money will be "harder" (more resilient in value) than the local money that preceded it.

Time preference
Interest rates are typically divided into money risk, asset/agent risk and the “risk-free cost of capital” (plus other transaction costs, which we can ignore). The “risk-free cost of capital” is better conceived of as pure time preference; what people are concerned about in any delay in use of income, which can differ across time and space. If people’s concern about delay is dominated by fears that they won’t have income later, time preference may even be positive—i.e. people become willing to pay to have access to their capital in the future.

In other words, interest rates should be treated as entirely an across-time price: thus if there was no delay risk or fears about future income, people would be indifferent to which time period they had access to their income (capital which was genuinely “risk free”--so no agent risk--would have no cost across time, apart from money risk). Though risk of death does, of course, give us a reason to be concerned about delay.

A nice short discussion of shifts in general risk since the medieval period is at William J Bernstein, The Societal Risk Premium. For a discussion of situation where pure time preference is positive (i.e. there is willingness accept negative returns in order to have future access to capital) see John Hempton's essay, The Chinese Kleptocracy Is Like Nothing In Human History.

As an aside, time preference should not be regarded as a single number, but as a continuum for agents across which the supply and demand for credit is matched--thus, if demand for credit sharply increases compared to supply, interest rates can be expected to rise, likely attracting new credit from people whose time preferences are now being covered and so are more willing to offer credit. (This can be understood as movement along a supply curve.) Conversely, if the demand for credit sharply decreases compared to supply, then interest rates can be expected to fall, reducing the number of people whose time preferences are covered and so are willing to offer credit. (This is also movement along a supply curve.)

If people increasingly judge that general delay risks are falling, their time preferences will shift, affecting their willingness to offer credit. (This can be understood as a shifting supply curve.) If such tendencies become entrenched, major effects can follow. Such as changing the arrangements of farming fields. Most peasant societies handle farming risk by dispersal across space (using scattered fields, thereby accepting reduced average production in order to reduce production variability from year to year). If interest rates fall, allowing much cheaper smoothing of income across time (either by increased use of less costly/risky storage or by use of--now significantly cheaper--credit), farmers tend to shift to concentrating fields, thereby increasing average production while accepting increased variability in production from year to year. That is, they move to managing variability across time rather than across space. Hence falling interest rates led to the enclosure movement (pdf).

So, the market for credit is not the market for money and interest rates are not a good basis for judging the stance of monetary policy.

Housing market fail
Then there is the Jordà, Schularick & Taylor paper's treatment of housing markets as if they are generic. A particularly egregious example is given at p.14:
Viewed as a natural experiment, the question is whether these differences in monetary policy treatment led to different outcomes in Ireland and Spain using Germany as control.
Paul Krugman famously divided US housing markets into the Zoned Zone and Flatland. Germany is Flatland, so it cannot be used as a control for any housing markets where the supply of land-for-housing is rationed (such as Ireland: by rationed I mean restricted in a way which significantly reduces the responsiveness of supply to increased demand, typically by various regulations).

Just as with money, where we need to distinguish between demand for money-as-asset and use of money to purchase goods and services, so we have to distinguish between demand for house-as-asset and use of housing--i.e. the demand for shelter. (Actually, as Matt Yglesias has pointed out, houses are large decaying structures, it is the land the housing is on which is the enduring asset.) The demand for shelter can be satisfied by buying or renting, a choice which will depend on the circumstances of particular people and the structure of a given housing market. The demand for house(land)-as-asset can only be satisfied by purchase.

So, if people think that land is a particularly good asset--because, for example, land-approved-for-housing is rationed, so has an entrenched tendency to increase in value faster than inflation--then the demand for house(land)-as-asset will be greater. A choice that will be definitely affected by interest rates, as low interest rates imply lower borrowing costs, so encourage more people to take out mortgages to enter the market for house(land)-as-asset, driving the price of houses even higher; if regulation or other constraints continues to inhibit supply responsiveness. So, easy credit can certainly be expected to have an upward effect on house(land)-as-asset prices in land rationed housing markets (or other housing markets with strong feedback effects). But, as discussed above, the price of credit and the price of money is not the same thing. Hence monetary policy having very little correlation with house price rises, but capital inflows having quite a strong correlation, as then Fed Chair Ben Bernanke pointed out in a 2010 speech.

What do we know about markets for assets in restricted supply? They tend to be more unstable (pdf). Hardly surprising, since there tends to be a positive feedback effect, where rising prices expected to continue of themselves encourage more demand for the asset. Particularly as there is experimental and empirical evidence that lack of experience in investors increases feedback effects; with house-buyers largely being inexperienced (or, at least narrow-experienced) investors, in that they rarely have experience of investing with other major assets, or engage in many house purchases; factors especially important when various housing markets had prolonged price-build ups.

So, using Germany--where supply of land-for-housing is responsive to demand--as a control for Ireland, where it was not, is a basic failure of analysis: an apples-with-oranges comparison. (In the case of Ireland, access to some areas was highly rationed, driving up prices, and to other areas was much less so, encouraging misallocation.) Supply and demand, they matter, really they do. Thus, the structure of particular markets matter, really they do.

What we appear to have here is data disease--"we have assembled all this lovely data, and it is much more pliable for our analysis if we just abstract across housing markets". But that is precisely what one cannot do--at least, not across housing markets with very different supply dynamics, for example.  The paper's failure to consider the difference between demand for shelter and the demand for house(land)-as-asset is quite clear:

The house is a bundle of the structure and the land used in its construction. An ideal index would capture the appreciation of the price of a standard, unchanging house which is hard to identify (p.5).
Monetary policy general: assets specific
Then there is the general versus particular problem. Monetary policy is a general phenomenon while "bubbles" (i.e. asset price surges and collapses) occur in specific asset markets. It is always a pertinent question, why did a surge-and-collapse occur in this asset market and not another? The US, for example, does not have a single housing market, it has hundreds, with wide diversity in levels of boom or bust.

Monetary policy also has a poor record when applied to asset "bubbles", including housing asset "bubbles". (Which, btw, is not a helpful way to think about housing price dynamics.) Lars Svensson noted that such "leaning against" policy can actually make debt-to-gdp ratios worse (pdf), by depressing income needed to service past debt more than it discourages future debt. Folk such as Richard Koo in Japan and Steve Keen in Australia typically apply Irving Fisher's Debt-Deflation Theory (pdf) of depressions in this unbalanced way--too focused on debt, not enough on income expectations (the deflation bit). There is no more effective way to produce disastrous debt dynamics than tight monetary policy driving down income expectations, thus spending, thus income, thus ability to service debt. Which is what happened in both the Great Depression and the Great Recession and continues to operate in much of the Eurozone.

The Jordà, Schularick & Taylor paper argues that the unified monetary policy of the Eurozone provides a situation for countries where the unified monetary policy was "too loose" to have credit booms through the "credit channel" of monetary policy (Pp11ff). First, as previously noted, it is surely more a matter of a unified monetary realm changing expected risks in various economies. Second, if it is a "monetary policy via credit channel" issue, one might expect that a similar process could operate in the US, since it is about as large as the Eurozone and also has a common monetary policy. Which then raises the issue of the paper's different treatment of the US versus the Eurozone--either treat the various US housing markets separately or treat the Eurozone as one big housing market, as the paper treats the US. Treating the US as one housing market and the Eurozone as a collection of them is worshipping far too much at the altar of national statistics rather than market dynamics.

Extending the former point, what the creation of the Eurozone did do was to eliminate money risk differentiations between economies as well as exchange rate risk: the consequent drop in nominal interest rates would be expected to increase demand for credit and the elimination of exchange rate risk increase the supply thereof in countries where both factors acted most strongly (i.e. the Mediterranean economies). But, as the paper points out, the effects on housing markets were not consistent--which again points to need to examine the structure of specific housing markets. A paper such as this (pdf) which provides a simple model for explaining housing price bubbles also (very sensibly) confines itself to policy recommendation which are specific to housing markets.

The Jordà, Schularick & Taylor paper holds that long-term interest rates are a good proxy for mortgage rates (p.21). Well yes, since mortgages are classic long-term financing and even more since mortgages have become such a large part of banking business (p.5). But, to invoke Milton Friedman once again, interest rates are set in a whole range of linked asset markets, so the question of why credit is drawn to a particular asset class still remains very relevant. Yes, monetary policy fundamentally affects general income expectations which will have a general tendency to push up (or down, depending on what income expectations it is generating) asset prices, but that is not even close to saying that monetary policy drives asset prices in anything close to a uniform way.

The failure to examine land use regulation in any systematic way also shows up in this paper bv Sebastian Dellepiane, Niamh Hardiman & Jon Las Heras on the housing boom-and-bust in Ireland and Spain. The paper notes that Portugal, Greece and Germany had very different experiences, but does not engage in any systematic examination of differences in land use regulation. (There is a passing reference to zoning in Ireland, and some hand-waving about de-regulation in Spain, but that is about it.)  The Dellepiane, Hardiman & Las Heras paper also engages in the unfortunate usage of "easy money" when they mean easy credit. On the other hand, the paper is clear that encouragement of "housing(land)-as-asset" was very much central to the difficulties in Spain and Ireland.

Either way, nice data and fancy maths do not warrant ignoring basic dynamics of supply and demand.

About history
If you are going to study the effect of monetary policy and monetary conditions on asset prices, then one really must familiarise oneself with C19th economic history. There you will find some truly spectacular asset booms and busts under a gold standard monetary regime: not generally regarded as conducive to "loose" monetary conditions. Yet one marked by generally low interest rates.

In particular, technological uncertainty (of which there was a great deal in the C19th) is more than enough on its own (pdf) to create asset booms and busts. But so will the supply of capital increasing in a way that it is "pushing on" investment opportunities; as per Andy Harless's analysis here.

Speaking of whom, he made an apposite comment about low discount rates and asset price volatility:
Low discount rates (which may not be quite the same thing as low market interest rates) do make assets hard to value (which may not be quite the same thing as causing bubbles), because they make asset values more sensitive to flows in the more distant future, which are harder to estimate. For example, if you assume a constant growth rate, as asset value is V = d/(r-g), where d is the current flow (“dividend”), g is the growth rate, and r is the discount rate. If r is only slightly higher than g, then a slight change in g will have a dramatic effect on V. So suppose r is low, and people happen to get a little too optimistic about g. It’s debatable whether this fits the definition of a bubble, but in any case it’s going to cause the price of the asset to skyrocket, and when the overoptimism fades, the asset price will collapse. (Of course the situation is symmetric. Maybe people rightly became optimistic about g, the asset price rightly skyrocketed, and then people happened, wrongly, to get just a little bit less optimistic, and the asset price collapsed. In this case there wasn’t a bubble per se, but the same volatility problem exists.)
To which monetary economist Scott Sumner responded:
Yes, asset prices might be more volatile, but that would not be because of monetary policy in any case. At least not for any extended period of time. Suppose rates were low, but inflation was below target. The Fed could raise rates, but that would drive inflation even lower. In that case either rates would fall again, or we’d go into deflation, and rates would fall a bit later. 
In any case, I see asset price volatility as being very different from asset price bubbles. Japanese stock prices have been very volatile over the past 20 years, but there has been no Japanese stock bubbles over the last 20 years. So it’s not clear that this sort of asset price volatility is a problem.
So, we are not talking about simple asset price volatility and housing prices were not notably affected by technological change (except, perhaps, changes in financing technology).

Comments I also made on the same post are apposite.  If there is a prolonged period of rising incomes are not asset prices going to rise? Not merely from expectations of future incomes but such further magnified by (rising) savings pushing on investment. And if expectations of increased income increased asset prices why would not also positive expectations about an asset as a store of value? Whether for gold or approved-for-housing land.

Moreover, in a market economy with multiple interest rates and multiple assets, with assets being various bundles of income-expectation, store-of-value-expectation, and expected risks, how can one measily central bank rate be more important than any of the aforementioned? And, in a market economy with multiple interest rates and multiple assets, with assets being various bundles of income-expectation, store-of-value-expectation, and expected risks, how can one interest rate be “the” rate that balances planned savings and planned investment?

To which the response is, it can't. (A weakness of Austrian analysis is that, on one hand, it insists that the heterogeneity of capital is crucial, yet it also gives a single interest rate amazing pivotal power.) One has to look at the dynamics of different asset markets. (Another problem with Austrian business cycle analysis is that an implication is that downturns should hit industries in a sequence according to their different average investment cycles, rather than all at once, as actually happens.) And housing markets are not asset markets in an all-the-same-really sense, as specific housing markets are not equally driven by the demand for shelter; in large part because they are not subject to the same supply dynamics.

And no amount of data assembly and fancy mathematics gets around that. Just as it cannot get around interest rates not being good indicators of monetary policy: in particular, low interest rates are very much not a reliable indicator that monetary policy is "loose" or "easy".

Scott Sumner famously keeps reminding folk to never reason from a price change. Another way to put that is: always attend to supply AND demand (even if no one understands that no one understands supply and demand). And so attend regardless of much data you have assembled and how much fancy maths you can apply to it.

[Cross-posted at Skepticlawyer.]


  1. An even simpler way to think about 'the price of money' is that if you've been in the habit of going into a grocery store and selling a dollar for one loaf of bread, but one day there's a new price tag that reads, $2 for that loaf, then the price of $1 has dropped in half.

    In which case, the price of borrowing money (interest) will probably rise.

    1. No, the latter will rise only if it was a rate of change increase for loaves (and many other things) and not a one time jump. And interest is the price of credit, no reason to refer to money in the definition just because we live in a monetary exchange economy.

    2. Saturos is right and Patrick is wrong. Assuming the change in the price of bread (and money) was a one-time jump, then you'd expect the interest rate to be unchanged.

      To see why, consider what would happen if you wanted to borrow, not $1, but 1 loaf of bread. Suppose that before the jump, you would have to pay back 1.1 loaves (i.e. interest rate of 10%). After the jump, the "bread" (i.e. real) interest rate would be unchanged, still 10%, therefore the nominal interest rate would also be unchanged (as ex hypothesi there was no change in ongoing inflation expectations etc).