Sunday, June 30, 2013

5 Pocket Models of the Macroeconomy

Economists, like all academics, spend most of their time becoming experts on very specific topics. One economist might specialize in the effects of micro-incentives in third-world countries, another might focus on the applications of game theory to the organ donor market, and a third may spend years pondering the causes of a particular historical economic event. When asked about their area of expertise, they give highly detailed, properly qualified, and usually accurate answers. However, when time is short and the question is novel, as when a policymakers need to respond to new and therefore unique macroeconomic event, economists tend to rely on a handful of basic models that most undergraduates learn in Econ 101.

The particular model varies by economist. For example, Paul Krugman proudly proclaims the IS-LM to be his favorite mistress.  In his book A Term at the Fed, economist Lawrence Meyer says that he relied on the concept of the NAIRU to guide his recommendations as a Federal Reserve Board governor from 1996 to 2002.  However, because his contemporaries on the FOMC, the committee that raises and lowers US interest rates, used different concepts and models to form their opinions, they often came to different conclusions. I suspect that an accurate map of economic ideology could be drawn simply by looking at which "pocket model" economists reference when answering questions on the fly.

As the "About" section of this blog proudly proclaims, I'm a quack without a flock, which means that I use a quirky pocket model of my own creation, although its basic principles aren't too far off of the mainstream macro track. I call it "The Spaghetti Model." They say that "it takes a theory to beat a theory" and I also often think that "it takes beating a theory to make a theory," so before making pasta, I'll explain why I don't like using the other common pocket models. I'll cover four:

  1. Classical Principles  
  2. Keynesian AS-AD (Aggregate Supply-Aggregate Demand) 
  3. Neoclassical AS-AD (Aggregate Supply-Aggregate Demand) 
  4. IS-LM (Investment-Savings-Liquidity-Money)
The DSGE is not on the list, primarily because the DSGE is not a pocket model, and secondarily because another post on this blog and also the theory tab tangle with it directly. In the interest of time I've also neglected Austrian and Post Keynesian models, which are respectively right and left of the mainstream and rarely used by policymakers in power. 

1) Classical Principles

Briefly, classical economics proclaims that markets are always efficient, money is always neutral, prices are always flexible, unemployment is always voluntary, supply always creates its own demand, and government intervention is almost always a bad idea. Classical principles are pretty easy to dismiss:


Despite its general lack of explanatory power, many economists still quote the tenants of classical economics like verses from the bible.  One must admit that classical ideas are appealing. Green pieces of paper aren’t valuable, so why should the quantity of money have any affect on the real economy? People can always offer to work for less money, so how can there be unemployment, except by choice?  Classical economics got a second kick in the 1970s and 80s, when real business cycle modeling seemed for a time to support old ideas.  Finally, economists who don’t like the AS-AD or IS-LM sometimes keep classical principles in their pockets simply for lack of a lesser evil.   Although today applied mainly out ignorance or out of spite, classical economics still influences policy and cannot be ignored. 

2) The Keynesian AS-AD Model

Supply and demand first entered mainstream economics with Alfred Marshall’s 1890 textbook, Principles of Economics, but originally the two iconic curves were used to describe equilibrium only in individual markets.  In 1937 in The General Theory of Interest, Employment and Money John Maynard Keynes put forward a theory of sticky wages and effective demand that allowed economists to draw the macroeconomy on axes of quantity and price. If aggregate demand increased (though mechanisms described by the IS-LM model, to be discussed), he argued, total real output, total employment and the price level would also increase.

In 1958 the unassuming New Zealand born economist William Phillips published a paper showing an empirical tradeoff between unemployment and inflation in historical UK data. His findings fit the Keynesian model well; greater employment came at the cost of inflation, and vice versa.  Governments could push the economy right or left along the “Phillipscurve” by manipulating aggregate demand through monetary and fiscal policy:


The IS-LM, the AS-AD, and Phillips curve formed the foundation for macroeconomic policy in the 50s, 60s and early 70s, until the Neoclassical AS-AD model replaced it as the pocket model of choice in the late 70s. Because of it’s similarity to basic supply and demand, the Keynesian AS-AD model is often used to introduce students to the study of aggregate relationships, but is generally snubbed by serious policymakers, because of what happened next.  

3) The Neoclassical AS-AD Model

In the late 1970s the economy started behaving in a way that defied the predictions of the Keynesian AS-AD; both unemployment AND inflation began to rise. In search of a better pocket model, Policymakers turned to Milton Friedman, who in 1967 had pointed out that the mainstream macroeconomic framework was missing one essential concept: inflationary expectations.

Friedman argued that if the government announced policy intended to increase employment, producers would expect inflation, and would therefore raise prices without increasing output or employment. Instead of pushing the economy right along the Phillips curve, the entire curve would shift out. With so-called “adaptive expectations,” Friedman continued, unemployment always tends toward a natural rate, determined roughly by the frictional unemployment necessary to match employers to employees and also possibly by the artificial minimum wage. In other words, government policy alone cannot produce real growth.



Further, Friedman and his supporters argued that the only thing that could increase output in the long term was an exogenous shock to productivity. This neoclassical spin on the AS-AD allowed that because of price stickiness, unemployment could in the short term rise above or below this rate, but such imbalances would always eventually be corrected by changes in the price level. Only the natural rate of unemployment tied to the natural rate of output produced price stability, a quantity he dubbed the Non-Accelerating Inflation Rateof Unemployment, or NAIRU.

This probably violates Mankiew's copyright, but you can see his explanation on slides 14-34 of this powerpoint. 
Today the Neoclassical AS-AD, with the IS-LM, is one of the primary macroeconomic models taught to students, and probably the most generally accepted on this list. While I think that Milton’s incorporation of expectations was appropriate, in my eyes the Neoclassical AS-AD is still far from the perfect pocket model. Three reasons:
  1. It offers no explanation for aggregate demand, an omission that becomes acceptable if the model is taken in conjunction with the IS-LM, but the integration isn’t clean and I’d also rather have one working pocket model than two half-functional ones.
  2. It completely divorces technological progress from demand, which given a second’s thought is obviously incorrect. The Internet wasn’t handed down to us by benevolent exogenous gods, it was developed and funded by those who anticipated that consumer demand would produce massive profits. Ideally we should be able to model the interaction between business cycles and long-term growth in the same model.
  3. On a related point, the NAIRU is hard to accept, both morally and empirically. If the goal is to conquer unemployment, assuming a natural rate admits defeat before the battle. More importantly, a natural rate isn’t “naturally” discernable from the data. 

4) The IS-LM model

The IS-LM model was first formalized in 1937 by John Hicks in his paper "Mr. Keynes and the Classics: A Suggested Interpretation," and was again was based on ideas sketched during by John Maynard Keynes in the The General Theory of Interest, Employment and Money. With the Keynesian AS-AD and the Phillips curve, IS-LM was hot in the 40s, 50s, and 60s.  Today the IS-LM is taught to undergraduates but almost never applied in formal academic papers.

The IS-LM plots two curves on axes of output and interest rate.   The IS curve is downward sloping because when the interest rate is high, people tend to save more and spend and borrow less, and when the interest rate is low, people tend to borrow and spend more and save less. The LM curve is upward sloping because the demand for money, or liquidity preference, is positively correlated with output and negatively correlated with the interest rate.

The IS curve can be represented by this equation:


And the LM curve by this equation:


The IS-LM curve isn’t particularly easy to understand, and many explanations exist; Hicks starts from another place entirely, Krugman takes a useful angle here. In his standard economics textbook, Greg Mankiew the two curves from two other graphs, the Keynesian Cross and the Money Market:

Again violating Mankiew's copyright, check out slides 35-69.

The IS-LM is quite handy because it allows one to easily model the effect of fiscal policy and monetary policy. Taken in conjunction with the Keynesian AS-AD, it was thought to model long-term effects on output, but after adaptive expectations came into favor economists began to think of the IS-LM as only as a short-term model. (Although what “short-term” means is still up for debate).  Today, Paul Krugman models liquidity traps with the IS-LM; when equilibrium dips below the “zero lower bound” the economy is stuck at low output, and won’t get out until we can manage to move the IS curve out.

Krugs creative manipulation is actually pretty useful, although still weird 

Partially because it’s so oddly constructed, many people strongly dislike the IS-LM. Of the four models listed so far, the IS-LM is by far my favorite, but still, three problems:  
  1. It cannot model the effects of changes in productivity on output or aggregate demand and also has nothing to say about the price level. These omissions become acceptable if the model is taken in conjunction with the AS-AD, but again, I’d rather have one model. 
  2. It draws a bright line between illiquid interest-bearing assets and liquid non-interest bearing money, which may have been a legitimate assumption in Keynes’ day when financial markets were less developed; but in today's world, when consumers can cash in securities in five minutes with a smartphone app, does a  general theory of liquidity preference still make sense? It might; I'm not sure. (For the more wonkish reader: I'm cool with the precautionary motive, but not with the transactions motive, and I'm iffy about the causal power of the speculative motive). 
  3. In general it places too much importance mechanical relationships, and not enough on subtler shifts in expectations. You can correct for this by observing that changes in expectations shift the constant variables, but then again if you’re determined you can hack almost any model to show what you want. Take Krugman’s zero lower bound application. If the only way to illustrate current events with the IS-LM is by adding a quadrant that was never intended to be part of the model in the first place, maybe we can do better?


Another commonly cited problem with the IS-LM is that it implies that the central bank targets the money supply, not the interest rate directly, as most central banks have been doing for decades. David Romer of the University of Berkeley proposed an adjustment, which he labeled the IS-MP (Investment-Savings-Monetary Policy) model to better reflect this reality, which essentially replaces money market with a horizontal line determined by the central bank. I don’t think that this adjustment cuts at the heart of my problems with the IS-LM, and I also don't like how it pretends that the central bank can play god. (Because let's face it, sometimes Bernanke is just pushing on a string). 

5) The Spaghetti Model

To define The Spaghetti Model, I’ll start from two problems all of the models above share.

  1. First, none deal well with time. Even the Neoclassical AS-AD, which incorporates curves for both short term and long term supply, is at any one point only a snapshot of a single period.
  2. Second, it’s not always easy to translate the models into empirical data. Wouldn’t it be nice to build a model from what we have? The five most important economic aggregates we track today are probably:
    • Nominal GDP
    • Inflation
    • Unemployment
    • Monetary aggregates
    • Interest rates
Given these two considerations, let’s start by graphing Nominal GDP over time:  
Units of Account is more accurate and more internationally inclusive than dollars. 
 Next, let’s add in inflation. We could just graph inflation on the same chart, as is often done, but instead let’s use inflation to calculate Real GDP. Plotting Real GDP on the same axes as Nominal GDP requires us to pick some point in time where Nominal GDP = Real GDP, but because we’re concerned about relationships, this point is arbitrary. A positive rate of inflation increases the gap between the two curves, and deflation decreases it.
Next, unemployment. Macroeconomists often speak of a hypothetical quantity called “Potential GDP,” the output the economy could produce if it was at full employment. Potential Real GDP is usually thought to be determined by the level of technology and the capital stock, and is difficult, if not impossible, to measure. So let’s use the unemployment rate to define Potential GDP relative to real GDP:

Now we have three curves, but still really haven’t said anything about money. Here comes the twist. If we can plot a curve representing Potential Real GDP, why can’t we plot a curve representing Potential Nominal GDP, the maximum amount of money you could possibly spend in a period? Potential Nominal GDP clearly should have some relation to monetary aggregates, but it’s equally clearly a separate variable, since nearly all of the monetary aggregates we track are lower than Nominal GDP. Potential Nominal GDP in a period should be the money times the potential velocity of money. As we assume that the level of technology and existing capital stock determine Potential Real GDP, we can assume that technology and capital determine potential velocity, the speed at which economic value can be exchanged. If technology improves or the money supply increases, Potential Nominal GDP will also increase.

Unfortunately, because it’s a composite variable we can’t chart Potential Nominal GDP directly off of measures of monetary aggregates. Enter the interest rate. We can think of the interest rate, like the unemployment rate, as a pressure gauge; as demand drives Nominal GDP toward Potential Nominal GDP, the pressure increases and the interest rate goes up, and vice versa. Defining Potential Nominal GDP in this way also conveniently allows us to sidestep defining it directly, like using unemployment to define Potential Real GDP deflects ugly questions about the nature of “value.”

Now we have a full model approximated with four of the five variables above, Nominal GDP, inflation, unemployment and the interest rate:

Simple. Straight. Uncooked Spaghetti. 
Expectations and government policy move curves, and the other curves react through push and pull.

  • If a positive state of expectation, or a bull market, pulls up Nominal GDP, the increase in spending will in turn pull on Real GDP and Potential Real GDP.  
  • Because production responds slower to changes in expectation than spending, Real GDP usually increases less than Nominal GDP in a bull market, leading to a rise in the price level as well as a rise in output, like in the Keynesian AS-AD model.   
  • Potential Real GDP responds even slower to changes in expectation, and often productivity gains don’t arrive until after the bull market is over. 
  • Because Potential Nominal GDP is partially defined by technology and the capital stock, productivity gains cause both curves to increase in tandem. 
  • The two "Potentials" or “Capacities” roughly chart long term trends, where the other two curves chart business cycles. 
  • If expansionary monetary policy increases Potential Nominal GDP, that will pull on spending, but if that monetary expansion is accompanied by an increase in inflationary expectations, it will not pull on Real GDP or Potential Real GDP. 
With the Spaghetti model, we can model a number of common macroeconomic phenomena quite easily:
Why spaghetti? Well, if you actually graphed these four curves with historical data, it would probably look a lot like spaghetti:


 3 More things about I like about The Spaghetti Model:


  1. It has NGDP at its heart and so fits with a currently popular and I think useful macro fad.
  2. The Federal Reserve seeks to influence unemployment and inflation via the interest rate, and the model conveniently relies on all three.
  3. Little about the model is mechanical, there are no all-important but unintuitive constants like the Marginal Propensity to Consume or the NAIRU to define. Some might complain that The Spaghetti Model leaves too much to discretion, and they would be justified; you can plausibly argue for almost any degree of push and pull. But then again you can do the same with IS and LM, AS and AD shifts. You can also pull out the push-pull and actual-potential relationships into charts that resemble the IS-LM and AS-AD; I'm too lazy to do that here, and this post is already enormous, but you can check out the last section of the theory tab for some of that. 
Most of all, I like this model because I find it to be most useful; that’s why I keep it in my pocket.


I stole this image from a momblogger


1 comment:

  1. Fred Grap!

    Graph it! You can do it!

    I graphed the stuff of one of Maynard's theory of crossing lines and it didn't seem to hold up.

    ReplyDelete