Milton Friedman's Thermostat

If a house has a good thermostat, we should observe a strong negative correlation between the amount of oil burned in the furnace (M), and the outside temperature (V). But we should observe no correlation between the amount of oil burned in the furnace (M) and the inside temperature (P). And we should observe no correlation between the outside temperature (V) and the inside temperature (P).

An econometrician, observing the data, concludes that the amount of oil burned had no effect on the inside temperature. Neither did the outside temperature. The only effect of burning oil seemed to be that it reduced the outside temperature. An increase in M will cause a decline in V, and have no effect on P.

A second econometrician, observing the same data, concludes that causality runs in the opposite direction. The only effect of an increase in outside temperature is to reduce the amount of oil burned. An increase in V will cause a decline in M, and have no effect on P.

But both agree that M and V are irrelevant for P. They switch off the furnace, and stop wasting their money on oil.

This is based on Milton Friedman's thermostat analogy (pdf). The central bank is the thermostat. This post is just another way of saying what David Beckworth is saying about James Hamilton's post on velocity. And it's another way of saying what Adam P. is saying about Menzie Chinn's post on the monetary base. And it's another way of saying what I said about why there's so little good evidence to tell if fiscal or monetary policy work. And the basic argument pre-dates Milton Friedman. Some old Keynesians (Maurice Peston?) figured out the basics back in the early 1970's.

If a policymaker is looking at information I to set a control variable C to target some variable T, then deviations of T around the target T* should be uncorrelated with C and with anything in I. The policymaker's forecast errors T-T* must be uncorrelated with the control variable and anything else in his information set, if the policymaker has rational expectations.

It's much more general than fiscal or monetary policy.

It is no surprise, for example, that Phillips Curves look flat when there's an inflation targeting central bank. Almost everything should be flat, if the target variable is on the vertical axis, and the central bank is doing its job right. It's only when the econometrician has got data on the horizontal axis that the central bank didn't have, and couldn't forecast, that there's a chance that the curve won't be flat. The only thing the econometrician is estimating is the central bank's forecast errors.

Update. Yep. Peston, M.H. 1972. “The Correlation between Targets and Instruments.” Economica 29(156): 427–31. Bottom line of that paper: there ought not be any.

Comments

[Matt Young]

The banker measures past forecasting errors, things that didn't go according to plan, is what I get out of this, after putting the econometrician on the banker's staff. If we had perfect foresite, we wouldn't need money, folks would work inside a bartering network with perfect agreements for future exchanges, the banker is not needed.

When all this gets backed into a model, we need prior agreement among all economic agents that planning is deliberately imprecise. We have the economic equivalent of the Copenhagen interpretation, economies only work because we defocus our image of goods over time.

[reason]

"If we had perfect foresite, we wouldn't need money, folks would work inside a bartering network with perfect agreements for future exchanges, the banker is not needed."

No, transaction costs are also an issue. And who would run the infrastructure that the bartering network uses? But yes uncertainty makes a huge difference, and any model that ignores it is suspect.

[Matt Young]

Transaction costs are a hidden entanglement of Nick's argument. It is transaction costs that cause the need for deliberately blurring the quantity and timing of arrivals.

[Nick Rowe]

The point of this post isn't really about money. It's about any policymaker/controller trying to use any instrument to target anything. We don't really observe the system the policymaker is trying to control. What we observe is the system+controller.

[Frances Woolley]

Nick: "But we should observe no correlation between the amount of oil burned in the furnace (M) and the inside temperature (P). And we should observe no correlation between the outside temperature (V) and the inside temperature (P)."

The first of these statements is only true if the second one is. That is, if people set their thermostat higher in summer and lower in winter, then there will be an inverse correlation between oil burned and inside temperature.

And it's rational to set thermostats higher in the summer than in the winter. E.g. suppose my preferred temperature is 21 degrees Celsius (around 70 Fahrenheit), and my willingness to pay to have the thermostat at 21 rather than 20 is $2/day. If the cost of heating is any more than $2 per day, I'll keep the thermostat at 20.

(Although, having said that, I know someone whose house is colder in summer than in winter because she loves the cool of air conditioning and the warmth of a snug home in winter.)

[Adam P]

Yep, this is a good post Nick.

I wish you'd put even more stress on the fact that the data imply more than just that the Fed is stabilizing the eonomy but that they're actually doing a pretty good job (at least for most of the data period).

Also, the point that in the the data you'd end up with the negative correlation between outside temperature and oil burned is important though it does mean that your wording is a bit sloppy here:

"An increase in M will cause a decline in V, and have no effect on P."

Really it's the decline in V that causes the fed to react by increasing M so that P won't change.

[Nick Rowe]

Frances: yep. So it even looks as though an increase in the amount of oil will cause a decrease in inside temperature, using quarterly data!

"And it's rational to set thermostats higher in the summer than in the winter. E.g. suppose my preferred temperature is 21 degrees Celsius (around 70 Fahrenheit), and my willingness to pay to have the thermostat at 21 rather than 20 is $2/day. If the cost of heating is any more than $2 per day, I'll keep the thermostat at 20."

That depends. There was a really good paper (actually, a lovely paper) on this about 10 or 20 years back. JPE I think. Given the laws of physics on heat transfer in conduction, the marginal cost of an extra degree of inside temperature is independent of the outside temperature. So you don't vary the thermostat when the outside temperature changes. But once the outside temperature reaches the inside temperature, the MC drops to zero. And then goes negative if you switch on the AC. Your example is at that corner solution, where what you say is right.

Adam: Thanks!

"Really it's the decline in V that causes the fed to react by increasing M so that P won't change."

Yep. *We* know that. But the econometrician might think "An increase in M will cause a decline in V, and have no effect on P."

That was the point I was trying to make. But I clearly can't have been clear enough. Damn!

[Adam P]

probably should have said something like "An increase in M will appearcause a decline in V..."

[Adam P]

appear to...

[David Beckworth]

Great post Nick. I loved the analogy.

Did you see Arnold Kling's reply? http://econlog.econlib.org/archives/2010/12/money_substitut.html

[Matt Young]

No it is not about money, but it is about streams of boxes of stuff. My claim is that if loaves of bread were a predetermined flow, then the rest of the economy would jump to a determined state.

[Jon]

Of course a combination of both is possible. Kling is right in the sense that Germany was able to use M2 targeting very successfully but they also had a more regular and orderly financial sector.

[K]

Nick: "Given the laws of physics on heat transfer in conduction, the marginal cost of an extra degree of inside temperature is independent of the outside temperature."

Getting way OT to the actual point of the post here, but... Not if you use a heat pump.  The amount of energy required to transfer Q units of heat from heat bath 1 to heat bath 2 is Q(T_2-T_1)/T_2, where T is the absolute temperature.  Heating becomes enormously efficient when the temperature differences are small.  The additional conduction loss due to an
additional degree of inside temperature is, indeed, independent of the outside temperature.  But the same cannot be said for the marginal cost of producing that extra degree of warmth.  So the the limits of the laws of physics definitely support keeping a higher indoor temperature when it's warm outside.

[Sina Motamedi]

Very good.

[RSJ]

The point is well-taken -- if the CB is succeeding in controlling the economy, you wont be able to see the underlying relationship between control variables and the objective variables.

But that statement, in and of itself, doesn't mean that the CB is succeeding anymore than it means that there is no relationship.

Where the rubber hits the road is in Japan, in which the CB has failed to increase output or prices by adjusting M or i, or, more recently, in the U.S. Perhaps the last two decades of successful targeting were a fluke, effective only within a certain environment -- secularly falling interest rates allowed households to increase borrowing each time the CB cut rates, but did not require a comparable de-leveraging when the CB raised rates. That also required a willingness on the part of banks to extend more credit to households on better terms each time the OIR was lowered, with the understanding that the collateral securing those loans would not decline in value when the OIR was raised backed up. That allowed for an apparent short term fine-tuning of demand by decreasing and then increasing the OIR, even though investment is long term, and the business sector has never been very responsive to short term OIR movements.

Before rushing out to praise the effectiveness of central banks, let's see how effective the current control-objective relationship is when it is placed in a different environment, for the next 20 years.

[Nick Rowe]

K's example of the heat pump has an interesting implication. Since the Marginal Cost of raising the internal temperature by one degree is an increasing function of the inside temperature and decreasing function of the outside temperature (the MC curve slopes up, and an increase in the outside temperature shifts it down), we would expect to see people turn down the thermostat on very cold days, and turn it up on warmer days. And we would see a *negative* correlation between electricity used and the inside temperature of the house. So it would look like more electricity causes a reduction in temperature (and increase in M causes a reduction in P) in a simple regression. A multiple regression of P on M and V wouldn't work, because there would be perfect multicolinearity between M and V.

[Matt Young]

I caught that too Nick, it was a good point. This is going to be fundamental, most of the economy is decorrelating from that fixed cycle, the season. Even roads going over big mountains requires multi-decorrelations of gasoline flows to gas stations on either side, for example. Control systems would have to deal with known congested ports. We would expect that, but it does not detract from your concept.

[edeast]

re RSJ. I read a paper saying inflation targeting was a fluke, but it was econometric, so I wouldn't mind if someone validated the techniques. It's called testing the good luck hypothesis. He analyzes Canada. Federico Ravenna

[Cleanthes]

Excellent analogy and fun to extend:
It turns out the thermostat is sentient and has declared that a "great moderation" is achieved. The inhabitants no longer worry about P. Yeah, a window has broken and termites are eating at the supports, but the house remains cozy because the thermostat is so good. In fact, one of the householders has decided that the front door should be opened to bring P to the outside, at least as far as the front yard trees, because the squirrels are cold.

[el chief]

What is the sound of a circle jerk in an echo chamber?

[Too Much Fed]

Anybody care to define M?

[Too Much Fed]

Adam P said: ""An increase in M will cause a decline in V, and have no effect on P."

Really it's the decline in V that causes the fed to react by increasing M so that P won't change."

OK, but are P and V related somehow?

[RSJ]

Edeast,

That's an interesting paper, but it still assumes that the current DSGE framework is basically correct, and in that framework there is no distinction between long term and short term rates, and the CB is assumed to control "the" nominal rate.

I'm not claiming that inflation targeting has no effect, but rather that the theoretical basis of the effect is a false assumption that the rate against which investment/consumption decisions are made is the same rate that the CB controls.

But we know this isn't true, because first, capital gains and losses need to be taken into account when measuring the rate of return against which household consumption/investment decisions are made.

And in that case, a lowering of the short term rate might lead to capital appreciation, so that the economically relevant rate might go up, not down, when the short term rate is lowered.

Moreover, there is no reason to believe that an adjustment in short term rates would have an immediate effect on long term rates, and yet we try to use short term rate adjustments as real-time control of a system governed by long term rates.

How is that possible?

What all of this tells me is that the channel by which short term rates affect the economy is not understood, and because of this, the actual channel may stop working or may not work as expected.

If you look at the actual history of this control variable, you see that (in the U.S.), a diminishing returns to using rate cuts to stimulate the economy, signifying that this isn't a particularly effective means of managing effective demand.

During each of the contractions, rates had to be cut by more and more in order to stop the contraction: by 2% in 1990, by 3% in 2001, and by over 4% in 2008 (hitting the zero bound prior to the expansion). And in each case, employment recovered to the pre-contraction level after longer and longer periods. It took 48 months for employment to recover to the pre-contraction levels in the aftermath of the 1990 recession. In the 2001 cycle, employment never recovered to the pre-contraction levels, and the recovery period in the current recession is going to be a matter of at least a decade, given the current policy response.

What you see is diminishing returns to the effectiveness of nominal interest rate control as the overall nominal rate declines. So here again, the nominal and not the real rate is what counts in measuring the effectiveness of CB short term rate adjustments.

To the degree that short term nominal interest rates are a tool that can be used to control the economy, the tool seems to be of limited usefulness because each time rates are cut, they cannot then be raised back to their previous level during the expansion, and simultaneous to that, the response of the system decreases with the overall level of the nominal rates.

It's in this sense that I said that interest rate targeting was a "fluke", in that the era began with high nominal rates, leaving a lot of bullets in the gun to use. Each time there was a recession, some of those bullets were fired, and now we've expended all the bullets and are at the zero bound.

[Too Much Fed]

RSJ, what you are describing seems to me to be an aggregate supply shock and not an aggregate demand shock? Sound good?

[edeast]

Ok, thanks for commenting on it. Every comment I write is at the edge of my knowledge, so I have nothing useful to say on your second point yet, maybe someone else will. But all of it was interesting.

[Doc Merlin]

But the Philips curve isn't flat... its all over the place.

[RSJ]

Too Much Fed,

No, I am not saying that at all. What I'm saying is that rate hikes and rate cuts are not symmetric operations, and that the stimulus effect of cutting rates relies on the ability of non-financial balance sheets to expand from their current levels. That means that when balance sheets can't/wont expand any further, you can't stimulate the economy further by cutting rates, and you've used up all your bullets.

One way I think about it -- and it may not be the right way -- is that when rates are cut, two things happen. One, on the margin, more investment occurs (typically residential investment). But simultaneous to that, existing debt is re-financed to the lower rate without necessarily creating any investment in the NIPA sense. This contributes to financial asset holdings of households but not to national incomes.

And you cannot re-finance when rates are increased, you can only re-finance when they are cut.

Therefore the economic drag of increasing rates by 1% is greater than the economic benefit of decreasing rates by 1%.

As a result, if you use rate cuts and hikes to manage aggregate demand, then there is a tendency for rates to fall and for balance sheets to only increase.

Here I am talking about nominal rates.

Because the economic cost of increasing rates is different from the cost of decreasing rates, there cannot be a function Y = Y(r), or even I = I(r). The function has to depend on other variables: I = I(Balance sheet, r, etc.)

And because of this asymmetry, any attempts to rely only on, or even primarily on, interest rate adjustments to maintain aggregate demand will eventually run into either the zero bound or a balance sheet recession. It's only a matter of time; it can take 20 years, or 30 years, it all depends on what your starting interest rates and balance sheet position was, and how it quickly they both deteriorate. But ultimately monetary policy is not a long term viable tool to manage aggregate demand.

It may be that no tool is long term viable, and that if an economy is prone to being deficient in aggregate demand, that there is no good solution other than mass redistribution, more union power, or something else.

Maybe there is no good way of managing an economy and every approach runs out of bullets at some point. But my sense is that the monetarism triumphalism is no more justified during the "great moderation" as the Keynesian triumphalism in the post-war era. Both approaches work until they stop working. If you were just looking at national incomes -- e.g. at the aggregate income statement, then it does appear that the last 30 years were successful at moderating volatility. But if you look at aggregated balance sheets, then they were anything other than stable. From the point of view of balance sheets, e.g. non-financial debt, the period 1950-1980 was stable, whereas the period of 1980-2007 was marked by instability. I think economists just didn't see this because in general there isn't a lot of attention focused on balance sheets as opposed to income statements, and as I argued previously, national income is an inappropriate aggregation of household incomes.

[K]

RSJ: Nice. Nominal rates are like a ratchet. Nominal debt is the pin.

[Too Much Fed]

RSJ, hopefully reply tomorrow.

[Nick Rowe]

RSJ:

You are basing your whole theory of the ratchet on a peculiarity of US mortgages -- that all (AFAIK) US mortgages are open mortgages, which give the borrower the option to pay off the mortgage early. An option they will exercise if rates fall (and then re-borrow).

Most mortgages in most countries are not like that, nor are most loans in the US, AFAIK. So, if your ratchet theory were right, it would be applicable only to the US, and even there to only a subset of the loan market.

But exercising that option merely causes a redistribution of wealth from lender to borrower, compared to a closed mortgage.

[RSJ]

Yes, that's true, but I do think that there is a general argument. But right now it's just a vague hand-wavy argument.

I think this goes back to the same debate we keep having, over and over again, in which I say that loans create deposits so that growth in borrowing (for the purchase of assets) adds to aggregate household income, whereas you say that credit growth and contraction merely shifts income around and doesn't do anything for aggregate demand.

In the former case, when the CB cuts short term rates, it is first and foremost causing the yield curve to steepen -- much more so than causing long term rates to decline.

Both will occur to some extent, but the primary effect is a steeper yield curve, or more profitable lending.

Let's say that this encourages banks to lend more -- they can cut lending standards somewhat, since each loan is more profitable and therefore they can bear more losses per loan, so their pool of potential customers increases. At the same time, they can also offer more attractive rates. It's a bit of both, of course.

The additional net borrowing is going increase aggregate household incomes, causing household budget constraints to shift out, this makes it easier for firms to meet their (longer term) financial profit requirements, even if their autarky profits (e.g. viewed as real economic value add) are lower. So this is cheating -- the financial profits are goosed by debt growth and do not reflect the real underlying economic value add. But firms (and households) are committed to delivering financial profits, not economic value add. Corporate earnings have been growing faster than GDP for 30 years. Household net worth has been doing the same. There is no requirement that financial savings or asset values be somehow tied to the real capital stock.

But at the same time, the stimulative effects arise from growth in borrowing, rather than the level of borrowing. As soon as debt levels stabilize, then this acts as a drag on demand since the growth in borrowing is decreasing. When this happens, firms are not meeting their cost of capital again so they start to liquidate and the CB decides to steepen the curve some more up until output is restored to potential, i.e. until debt growth resumes to its previous level.

And that's all you need for a "general" ratchet argument that debt levels need to keep rising faster than GDP, if they start rising faster than GDP at the beginning, allowing unrealistic (financial) return requirements to be met for a period of time, even while the autarky return is lower. If you start that process, and if the CB has a mission to keep cutting rates until output returns to trend, then you will stay on that route -- we can call this the bubble route.

OK, you will probably reject this argument because you wont agree that an increase in net borrowing (by households for the purchase of assets) causes household incomes to increase. But I (hope) that if you grant this point then the argument would go through.

None of the above is a "proof", though, but I hope it makes sense.

[Too Much Fed]

I said: "RSJ, what you are describing seems to me to be an aggregate supply shock and not an aggregate demand shock? Sound good?"

RSJ said: "That means that when balance sheets can't/wont expand any further, you can't stimulate the economy further by cutting rates, and you've used up all your bullets."

It seems to me that expanding balance sheets (with what medium of exchange?) should eventually cause price inflation if an economy has an aggregate demand shock.

What happens if there is a continuous, aggregate supply shock?

[Too Much Fed]

RSJ said: "And because of this asymmetry, any attempts to rely only on, or even primarily on, interest rate adjustments to maintain aggregate demand will eventually run into either the zero bound or a balance sheet recession."

Only if supply is growing faster than demand until supply is greater than demand?

[Too Much Fed]

"From the point of view of balance sheets, e.g. non-financial debt, the period 1950-1980 was stable, whereas the period of 1980-2007 was marked by instability. I think economists just didn't see this because in general there isn't a lot of attention focused on balance sheets as opposed to income statements, and as I argued previously, national income is an inappropriate aggregation of household incomes."

I believe that if attention is focused on balance sheets (I'd rather call them budgets), people will eventually have to look at currency denominated debt levels. Some people might even come to the conclusion that the solution to too much currency denominated debt is not more currency denominated debt.

[Too Much Fed]

RSJ said: "I think this goes back to the same debate we keep having, over and over again, in which I say that loans create deposits so that growth in borrowing (for the purchase of assets) adds to aggregate household income, whereas you say that credit growth and contraction merely shifts income around and doesn't do anything for aggregate demand."

I'm not sure I would say household income. I'll say it this way. If I haven't saved up enough medium of exchange to buy a car, I can can borrow to purchase it if my budget allows. This act of borrowing brings future demand to the present and then "debt enslaves" me to make the interest and principal payments. Basically, the central bank can then attempt to control my borrowing thru the fed funds rate and the markup by the bank on the loan.