Against models: The Good, the Bad, and the Ugly

I got to thinking about macro models while reading a recent Mark Thoma post on the debate between those who defend modern macro models, and those like Krugman and DeLong who wish to move away from highly abstract and unrealistic models (new Keynesian or new classical)  and toward something more realistic.  I’m not quite sure what that is, but I believe they think it would be something that incorporates insights from people like Keynes and Minsky, and something able to deal realistically with policy options at the zero rate bound.

I know this is going to sound arrogant coming from a pipsqueak at Bentley, but I think both sides are wasting their time.  I think we need to move away from macro models, or at least macro models of the type that are the primary subject of dispute.  I can think of at least three types of macro models:

1.  Models that attempt to explain in very simple terms how key macro variables get determined; IS/LM, AS/AD, money supply and demand, loanable funds and interest rates, etc.

2.  Models that try to show the relationship between changes in macro variables like the price level and output, and unobservable changes in the social welfare function (such as the menu cost of inflation, or involuntary unemployment.)

3.  Models that attempt to predict movements in the key macro variables, and make conditional predictions based on various settings of monetary and fiscal policy levers.

The first type of model is fine, as a pedagogical device.  The second are useful in principle, but in practice we mostly go with the gut.  We assume that demand policy probably doesn’t affect the long run rate of real growth, so we try to reduce the amplitude of the business cycle.  We also believe that inflation should be predictable.  Many economists believe that a stable price level minimizes the welfare cost of inflation, almost all believe the optimal inflation rate in the US is between about minus 2% and plus 4%.  We really don’t have any way of showing who’s right, although the answer might depend to some extent how well we deal with the third problem.

The debate is focused on the various structural models of the economy.  What happens if we cut the fed funds rate by 1/2%, or increase the budget deficit by $300 billion?  We expect answers from our models, and the more Keynesian models generally give slightly different answers than the so-called “freshwater” models.

What I find so odd about this debate is that modern macro theory is often assumed to generate some sort of highly technical structural model featuring rational expectations.  But that’s not at all what modern macro theory implies.  Modern macro theory implies that policymakers should get the optimal forecast of the policy goal variables, conditional on various policy settings.  And modern macro theory suggests that the best way to do that is to create and subsidize futures markets that trade contracts linked to variables being targeted by policymakers.  Indeed in the case of monetary policy, money should be convertible into those contracts.  And since the price of those contracts presumably has no zero bound, there is no case for fiscal stabilization policy.

You might object that if we don’t build these models, then the markets will lack the sort of information required to make intelligent forecasts.  I suppose that’s possible, but I have two problems with that argument.  First, there will always be people trying to model the economy, so I’m not worried about that.  I’m not even opposed to having the government fund a couple hundred model-builders at the Fed on the grounds that the information it would provide is a public good.  What I oppose is the Fed actually using the information from their economic research unit.  Let the markets use that information, if they think it’s useful.

Will the markets think it’s useful?  Ask yourself this question:  In the first 10 days of October 2008, when markets rapidly scaled back NGDP growth forecasts as they received extremely bearish reports regarding AD all over the world, and saw the Fed policy seemingly helpless to arrest the decline, which macro model were markets looking at?  Don’t these models usually put in lagged macro data of various sorts, the sort of data that’s reported monthly or quarterly?  What sort of data would have caused such bearish sentiment that stock prices fell 23% in 10 days?  I suppose you could put in market data measured in real time, but if you are going to do that, why not skip the middleman?  Just create a market in the variable you care about, don’t try to infer NGDP growth by putting other real-time market indicators into your model.

So here’s how I see things:

The Good:  Create a NGDP market and stop trying to outguess the markets.  If you don’t like NGDP, create separate price level and RGDP futures markets, and stabilize a weighted average of the two contracts along the desired growth trajectory.

The Bad:  Try to create a structural model of the economy, under the assumption that the public sees the world the way you do.  I.e., if you think X causes Y, don’t assume the public believes that X doesn’t cause Y.  In other words, use consistent expectations (unfortunately these are misleadingly called rational expectations–it has nothing to do with rationality.)  Input data and predict.

The Ugly:  Create a model that assumes you are smarter than the public.  Also assume that monetary policymakers are really dumb, and don’t know how to do policy at the zero bound even though your model says they can do monetary stimulus at the zero bound.  But fiscal policymakers are wise, and free of all political influence.  Once all these assumptions are built in, you are free to indulge your wildest left-wing fantasies.  All the laws of economics go out the window.  No more opportunity costs—build pyramids in the desert, or high speed rail between Tampa and Orlando.  Payroll tax cuts cost jobs, but higher minimum wages actually create jobs.  Protectionism is also great, especially when directed against oriental people people who live in high-saving cultures.

I suppose that in the debate between the bad and the ugly I should support my freshwater alma mater.  But I can’t even get up enough enthusiasm to enter the fray.  Once you start thinking in terms of NGDP futures targeting, everything else seems pretty pointless.

PS.  Don’t tell me markets are irrational.  The point is to stabilize NGDP expectations; according to modern macro theory (Woodford, etc.) unstable expectations of future NGDP causes unstable current NGDP.

Update 5/28/10:  Nick Rowe has a related post on this issue.  I agree with his commenters that there is an important distinction between unconventional forecasts (aka witchcraft) and making conditional forecasts based on alternative policy choices.  But even the latter are inferior to futures markets.  After I did this post I realized that I should not have implied that all policy issues revolve around NGDP—I was thinking of demand-side policies.  It would be considerably harder to set up prediction markets to evaluate supply-side policies, but nevertheless there is no reason in principle that it couldn’t be done.  Interested readers should look at Robin Hanson’s work on “futarchy.”


Tags:

 
 
 

11 Responses to “Against models: The Good, the Bad, and the Ugly”

  1. Gravatar of Lorenzo from Oz Lorenzo from Oz
    27. May 2010 at 21:54

    Nice and clear (certainly clearer than the post you are responding to) and goes with my prejudices, so I like the post 🙂

    There also seems to be a deeper misunderstanding around about computer models in general: they test your premises and measurements, they are not actually evidence for how the world is apart from that. This may seem an obvious point, but it is clearly not.

  2. Gravatar of scott sumner scott sumner
    28. May 2010 at 04:20

    Lorenzo, That’s a very good point. Of course there is a deep philosophical question as to whether any model shows how the world actually is, or if they are merely useful for some purpose. As a philosophical pragmatist I take the view that models don’t literally describe reality, but rather are more or less useful tools. My view is not widely held with respect to physics models such as E=MC2 (although I think it even applies there) but I think it’s easier to see in macroeconomics. If we went to NGDP futures targeting, then it would essentially take nominal shocks off the table. In that case all economists (including Keynesians) would become RBC economists. It’s not that the RBC model would suddenly be “objectively true” and the Keynesian model “objectively false” rather the Keynesian model would have become (even more completely) useless.

  3. Gravatar of Mike Sandifer Mike Sandifer
    28. May 2010 at 04:24

    For what it’s worth, I’ve long been skeptical about macroeconomic models. I have this sense that even popular supply and demand models are wrong.

    I also sense that I’m likely wrong, given my severe ignorance on the subject.

  4. Gravatar of Cdn Expat Cdn Expat
    28. May 2010 at 05:39

    That was fun.

    You know, though, that rational expectations triumphed as a modeling assumption not because it was true but because invoking the assumption took those free parameters off the table that you’re referring to under “the ugly.” Lots of room for policy if you assume policymakers (or their model building geeks) are omniscient and private agents share the same congitive capabilities of a cow staring at a passing train.

    I’m not sure that creating markets in, say, nominal income, obviates problems the way you suggest. There already are fed funds futures markets but that doesn’t mean that the fed needn’t worry about how to choose a conditional path for the funds rate or worry about what it is private agents are thinking about when they expect quite a different path than does the Fed itself.

  5. Gravatar of Indy Indy
    28. May 2010 at 06:09

    My problem with the models is that every time you add another variable it becomes increasingly possible to econometric-ally fit the existing data regardless of the “fundamental accuracy” of the model.

    I suppose I mean “accuracy” in the pragmatic hard-science sense – that is – we “understand” something when we can routinely make accurate forecasts and predictions, and it doesn’t matter how “congruent” the mechanics of the model simulations are with the actual dynamics of the universe so long as they yield more or less equivalent equivalent in the factors we are trying to measure.

    As you start to add more and more variables though – you being to get specious results – results that look like they plausibly model add past data points but have limited prediction power.

    It’s unavoidable, but think of the ways such a phenomenon can be abused. If you have a basic theoretical framework for how the forces and flows in the system function, and you try to support or validate your theory based on how well it fits the past data – a radical new theorist can come along, add another variable to the mix and claim, “Aha, but this new model with this extra variable also fits all the known data, but implies a completely different theory. The two theories agree, coincidentally, for ‘normal’ circumstances, but my extra-variable-theory predicts that everything flips upside-down in this new special circumstance of ‘zero-bound’ (or something like that).

    I think Feynman once gave a lecture about the fine-structure constant – and how some Scientists had claimed to discover a very-close approximation in terms of some fundamental mathematical constants that they thought might lead us in the direction of where the number came from. They used some combination of e to pi and the square root of two and all that.

    And Feynman said something like, “That’s all well and good, but you know, there’s a mathematical theorem which tells us you can always get arbitrarily close to any number with a surprisingly small number of fundamental constants and arithmetic operations. You could change one operation and swap the square root of two for the cube root of three, and the result would only be different in the tenth decimal place and we would be no closer to knowing the origin of this feature of our universe.”

    So I do wonder then if it’s not simply impossible to judge between models based on different theories which agree in normal times but diverse in extraordinary circumstances. If it is impossible – then we should indeed abandon the effort as irresolvable and focus on things like NGDP targeting.

  6. Gravatar of Brendan Brendan
    28. May 2010 at 07:30

    “I know this is going to sound arrogant coming from a pipsqueak at Bentley”

    For predicting macro-variables:

    You recommend utilizing the aggregate views of tens of thousands of CFA’s, economists and professional money managers.

    And the opposing view point believes that the idea of a small group of individuals building macro models is sound. But that it’s been screwed up until now cause everyone is super dumb- except them.

    Who is arrogant?

    Great post as usual Scott. I seem to remember you claiming that you were a poor writer, but you’re very wrong. Maybe I like it cause I don’t know how to use commas either.

  7. Gravatar of Gregor Bush Gregor Bush
    28. May 2010 at 07:51

    Scott,
    Given that NGDP futures are unlikely to be adopted imminently, as a transition step, do you support the Fed publishing its own forecasts of the path of the fed funds rate along with its forecasts of growth, inflation and the unemployment rate in a similar fashion to the Ritsbank? This at least gives market participants a clearer reading about how the Fed sees the economy and the relationship between its objectives and its conventional tool. The Fed can then monitor the equity and TIPS markets and gauge whether or not its own projection of “appropriate monetary policy” is on track or is causing it to miss its target.

    Secondly, why does the debate about more aggressive monetary stimulus always turn into a debate about setting a higher inflation target? A two percent target will work just fine if you are explicit and committed to hitting it. The debate with Krugman, Delong, Summers, Richard Koo, ect should be about the actions/announcements necessary to hit the target, not about the target itself. Advocates of more fiscal stimulus are taking the bizarre position that, absent fiscal action, deflation is imminent but that any attempt to use monetary policy will necessitate higher-than-optimal inflation (or in Koo’s case, hyperinflation). No, it won’t. It will necessitate 2%.

    One thing that amazes me is that the Fed, by showing its long-run forecast of inflation in the minutes, essentially states that its inflation target is 1.75% on core PCE (which translates to almost exactly 2% on core CPI). It then shows its forecasts of fourth quarter year-over-year core PCE inflation of 1.0%, 1.3%, and 1.4% for 2010, 2011, and 2012 respectively (the mean of the “central tendency”). It also shows a range for 2012 Q4 of 0.6% to 2.2%.

    The message is clear: “Monetary policy has been too tight over the past six quarters which is why inflation has come in well below our target. And we plan on keeping policy too tight over the next ten quarters and likely quite a bit longer. We view missing our target by 1.2% to the downside and 0.5% to the upside two and a half years from now as equally balanced outcomes.”

    It’s right there in black and white. How can anyone not look at this and say: “My god! Sumner was right!”

    If Congress was smart (I know, I know), they would change the Fed’s mandate to a hard 2% inflation target. The Fed should have to answer to Congress whenever core inflation strays outside of a 1.5% to 2.5% band (like it has right now). In the current episode, the fuzzy “dual mandate” is ironically resulting in weaker rather than stronger employment growth.

    Keep up the good work. Looking forward to the book,

    Gregor

  8. Gravatar of Joe Calhoun Joe Calhoun
    28. May 2010 at 08:21

    Scott,

    I agree with your take on macro models but I want to zero in on one statement:

    “We assume that demand policy probably doesn’t affect the long run rate of real growth, so we try to reduce the amplitude of the business cycle.”

    As a reformed engineer, I often use the amplitude/wavelength analogy to explain economic policy. I tell people that we have the ability to reduce the amplitude of the economic cycle but the cost/benefit of doing so is increased wavelength (recessions are shallower but last longer; the fall in residential real estate prices is mitigated but clearing the inventory takes longer, etc.). For the last several decades (at least) it has been primarily the function of monetary policy to reduce that amplitude and it seems to me that the result was the so called “great moderation”. I wonder though if we aren’t now discovering there is a long term cost to reducing the amplitude of the business cycle. Each time the Fed stops the recession short of where it would have gone naturally, some excess (debt) is not purged from the system and doing this repeatedly allows these excesses (debts) to accumulate until we reach a point like the one we reached in 2008. In other words, what if this basic assumption we make – that reducing the amplitude of the business cycle is beneficial – is not correct?

  9. Gravatar of Ted Ted
    28. May 2010 at 09:51

    Firstly, I don’t have a lot to say regarding models, I agree with a lot of what you said. I think your underrate structural models usefulness, but I largely agree with your sentiment. So, I just have a nitpick and a question.

    First, my nitpick. You wrote:

    “Also assume that monetary policymakers are really dumb, and don’t know how to do policy at the zero bound even though your model says they can do monetary stimulus at the zero bound.  But fiscal policymakers are wise, and free of all political influence.”

    I don’t think political influence matters for fiscal stimulus since politicians love to spend money. My understanding of the channel through which fiscal stimulus works is by bringing down the real interest rate when the nominal interest rate is zero. When governments run large budget deficits, this creates an incentive for them to create inflation in the future, and so the private sector expects future inflation to help ease the debt. Thus, it would seem, large budget deficits via fiscal stimulus seems to be just a commitment device to future inflation. The way I read it, it doesn’t seem to matter how the money is spent. You could apparently just open up a giant fire pit and start pouring U.S. dollars into and it would have virtually the same effect. Although, my understanding also is, that if the monetary authority is not coordinating with the fiscal authority and doesn’t intend to allow higher inflation to deal with the budget consequences of fiscal stimulus then the effect of fiscal stimulus becomes very small. I certainty don’t believe the Fed is in any way committed to create future inflation to deal with the budget consequences of fiscal stimulus, I expect tax hikes to be preferred route. To me, this helps to explain why the fiscal stimulus have probably been ineffective. But so then, I don’t think it really matters about political influence since you are just trying to run up the debt to create future inflation expectations – and politicians love spending money.

    Also, I’m curious about one thing randomly about NGDP futures targeting. More often than not the futures market is efficient in the long-run (with some exceptions), but there is also evidence that quite a few futures markets are inefficient in the short-run. Considering that, to my knowledge, we do not understand why one futures market exhibits short-run inefficiency while another does not, should we be more cautious about NGDP futures targeting when are attempting to make real-time decisions? Now perhaps it is these short-run inefficiencies may not exist in an NGDP futures market, or perhaps they are small enough that the benefits of NGDP futures targeting are still greater than alternatives. But, I’m curious, do you believe this could be a problem with your proposal?

  10. Gravatar of scott sumner scott sumner
    29. May 2010 at 05:47

    Mike, S&D models aren’t perfect, but they are reasonable approximations of many markets.

    CDn Expat, The difference is that NGDP is the goal, whereas fed funds futures are just a tool to achieve the goal. NGDP futures have much more potential. Last year I did a post called “Find the flaw in NGDP futures targeting” which discusses the advantages of this scheme.

    Indy, Those are very good points. In addition, we have one post-WWII data set that everyone uses, so they build the model to fit the data.

    Brendan, Thanks.

    Gregor, I don’t think we need fed funds forecasts. If they won’t do NGDP, then they should publish both their goal for the core price level in 2 years, and their forecast of the core price level in 2 years, and the TIPS forecast of the price level in two years, and explain any discrepancies (all three should agree if policy is on target.)

    You said;

    “One thing that amazes me is that the Fed, by showing its long-run forecast of inflation in the minutes, essentially states that its inflation target is 1.75% on core PCE (which translates to almost exactly 2% on core CPI). It then shows its forecasts of fourth quarter year-over-year core PCE inflation of 1.0%, 1.3%, and 1.4% for 2010, 2011, and 2012 respectively (the mean of the “central tendency”). It also shows a range for 2012 Q4 of 0.6% to 2.2%.
    The message is clear: “Monetary policy has been too tight over the past six quarters which is why inflation has come in well below our target. And we plan on keeping policy too tight over the next ten quarters and likely quite a bit longer. We view missing our target by 1.2% to the downside and 0.5% to the upside two and a half years from now as equally balanced outcomes.”
    It’s right there in black and white. How can anyone not look at this and say: “My god! Sumner was right!””

    Great observations, especially the last part. 🙂

    Joe, I don’t agree with that view (which seems kind of Austrian.) I don’t think there is such a thing as a natural progression of a recession. Government determines AD, for better or worse. Nothing that happened to NGDP is natural. RGDP is another story. If the government did nothing to interest rates in a recession, the price level would fall all the way to zero—hyperdeflation–because real rates would keep rising.

    Ted, You said;

    “My understanding of the channel through which fiscal stimulus works is by bringing down the real interest rate when the nominal interest rate is zero.”

    I’m not saying this is impossible, but the reverse is usually assumed.

    I agree that the Fed won’t monetize the deficit, and hence that fiscal stimulus won’t work.

    I don’t think the inefficiencies would be big enough to cause macroeconomic problems. Thus I don’t think the market price of NGDP futures would deviate from the market forecast by more than 1%. I also discussed allowing the Fed itself to intervene if it saw problems with inefficiency, manipulation, etc. It would still be a useful tool, even with the Fed participating to some extent. It would still avoid the gross errors like late 2008.

    Last year I did a post called “Find the flaw in NGDP futures targeting” which discusses the advantages of this scheme.

  11. Gravatar of A practical critique of NGDP targeting | Basil Halperin A practical critique of NGDP targeting | Basil Halperin
    1. January 2015 at 03:04

    […] necessitating a structural model of the economy disturbs me, for a successful such model – as Sumner persuasively argues – will never be built. The prospect that NGDP targeting might collapse in the face of rapidly […]

Leave a Reply