Pages

Tuesday 17 September 2013

On the Role of Models



In the wake of Paul Krugman's post on Wynne Godley and hydraulic modelling, I've had a number of discussions, including this one with Phil Pilkington, on the role of modelling in economics.

If you've read a few of my posts, you will know that I use models a lot.  This in part a reflection of my laziness - I'm not very good at reading anyone else's stuff, if it extends to more than a page without equations in it.  I've always preferred to work things out for myself - an approach that has proved very successful for me in various aspects of life.  If I can't figure something out in my head, I scribble down a few diagrams or a balance sheet, and if that's not enough I build a little model.

These sort of models are intended to help me develop my own ideas about how such things work.  Ideally they should be stripped of everything other than the point I want to look at - as simple as possible whilst retaining the complexity to push my understanding a bit further.  The models I have included in some of my posts are examples of this.  They show things where I have a sense of how it works, but I need to see it in action to really get my head round it.

The type of model is driven by the issue I am concerned with.  But on the whole, I like models  that I can relate to the things we observe in a real economy.  I like to think that I could take any of these models and, maybe with a little tweaking, put some realistic numbers on them.  I wouldn't expect to get good forecasts or anything by doing that - it would all be about understanding how it really works.  Or doesn't.  Sometimes the conclusion is not what you expected.

I also look at more detailed models.  Again, I have on this blog some details on my UK macro model.  The purpose here is different, but not a lot so.  It is still about understanding.  A lot of it is simply the learning that comes through construction of a model.  Often simply having to organise and make sense of the data reveals important insights.  Otherwise the benefit comes from running experiments with the model.  With simple theoretical models, we are trying to understand a very specific mechanic.  With bigger, more complete models, the purpose is more general.  We are testing our intuitions, to see if everything works to together as we believe.  Often, this will draw out interesting effects that would have been hard to spot otherwise.  It's all about having a tool to aid our thinking - a grand version of a supply and demand diagram.

However, I think it is very important to recognise the limits to what models can do.  It is easy to get seduced into thinking that a model is some kind or oracle.  This is a mistake.  Any model is necessarily a huge simplification.  The results depend critically on the assumptions made.  However complex and detailed they are, all they really reflect is the theories of the modeller.

This doesn't invalidate the benefits I have talked about, but it means we must be careful how we use them.  They can help inform and quantify our judgements, but that is all.  If we don't understand the results, they are useless.  The model is not revealing any new truth, it is simply reflecting our own ideas, helping us to visualise how a massively complex system fits together.

I appreciate that there are economists have no interest in models.  That's fine - many of them I have great respect for, and the profession benefits greatly from having people take different approaches.  For me, the use of models is invaluable.  One of the things that most impressed me about Wynne Godley was his ability to combine economic theory with a deep insight into the real data.  His work in empirical modelling was key to that and it is the reason I do economics the way I do.

14 comments:

  1. Nick,

    I totally agree. I always say that I think that models are very useful didactically -- and that appears to be what you're talking about here.

    The question then becomes: what sort of models are didactically useful? And I would say, one's that try to emulate real-world processes as Godley's (and Lavoie's) do.

    So, the observation that (I hope) comes out of this are twofold: (1) Models are basically teaching -- and self-teaching -- tools and probably will not do very well if applied empirically to produce arbitrary numbers. (2) They should, in contrast to DSGE/New Classical/whatever models, mirror reality as closely as possible.

    Phil

    ReplyDelete
    Replies
    1. Yes. I would be like to open on different types of model, but on the whole I find it difficult to relate the DSGE models to the things I observe, so they don't really work for me.

      Delete
  2. Nice post. As an applied mathematician, I have a bias towards mathematical models, although you need to be skeptical about mathematics. My impression is that far too many people learn some techniques, then apply them as "best practice" without really understanding what they mean.

    But without some kind of mathematical model behind your logic, it is unclear how you can evaluate policy choices. How big a stimulus did the U.S. need in 2009? One billion dollars? One trillion? Without some sort of model in mind, you have no way of evaluating the decision.

    ReplyDelete
    Replies
    1. Thanks.

      I was reading through some stuff on the Cambridge Alphametrics Model (CAM) of the world economy yesterday and I came across this bit:

      "The model provides a tool for investigation and learning about behaviour of the world economy and interrelationships between events in different regions through exploration of historical data, construction and modification of core variables and equations of the model and study of projections or simulations of future policies and outcomes. To the extent that the model is in some sense realistic, such activities give model users a clearer sense of magnitudes and
      rates of change of different phenomena and provide indications of the likely outcome of different policy interventions and norms."

      from http://www.augurproject.eu/IMG/pdf/D1-1_Introduction_to_the_CAM_databank_and_model.pdf

      I like this. It seems to chime with what I wrote in this post.

      One of the people behind CAM is Francis Cripps, who was worked with Wynne Godley on the old CEPG model and co-authored with him the book "Macroeconomics".

      Delete
  3. A made a model/simulation for hyperinflation. I have not really gotten any feedback on the model/simulation. The few comments on my post were not really about the model. Do you have any advice on how I could get feedback on my model/simulation?

    http://howfiatdies.blogspot.com/2013/03/simulating-hyperinflation.html

    ReplyDelete
    Replies
    1. I'm not really sure, Vincent.

      I would say one thing though. On your blog, I could see how to run the simulation, but I couldn't see anywhere where it set out how it worked. For me that's not very interesting. As per my post above, I think it is all about seeing how the result flows from the assumptions. To do that, I need to see the equations. But maybe that's just me.

      Delete
    2. I looked at it quickly, and I think this illustrates one problem with didactic models - a mathematical model just simulates what you have embedded as assumptions. Your model shows how a hyperinflation is possible, but the question remains: do the embedded assumptions make sense?

      If one approaches economics as a purely literary discipline, there is no good way to say that one set of assumptions is better than another.

      Your model appears to rely on two key assumptions: interest rates will rise as a result of increased debt, and that indexation is perfect throughout the economy.

      I showed recently on my blog (www.bondeconomics.com) that rising government debt/GDP ratios are associated with lower nominal interest rates. And indexation is not exactly a major issue in the major economies right now. For example, the spike in inflation due to oil prices in 2008 helped cause a collapse in sales volumes, not higher wages. Thus, there are reasons to question your embedded assumptions. But to do that, you need to look at the historical data.

      Delete
    3. If you click on each node you can see the equation off to the right. You can even click on "Clone Insight" and then you can have your own copy of the whole simulation where you can change the formulas. To change them click on the "=" on the node.

      Delete
    4. In that simulation the interest rates go down at the start and then stay down. I am trying to simulate a central bank going for a zero interest rate policy and showing how that can cause deflation as interest rates go down.

      The velocity of money is a function of the interest rate. So as interest rates go down the velocity of money goes down and the price level can go down too.

      Oh, I guess you have to click on the "=" on the node to see all the detail of the formula.

      Delete
    5. Not sure what you mean by "Indexation is perfect". I use the equation of exchange and do assume that is always true. With values for GNP, money supply, and velocity of money, I can then calculate a price level. I have yet to find an economist that thought the equation of exchange did not always hold.

      Delete
    6. I needed to clone the simulation to be able to see the equations properly.

      What I mean by indexation: your model is based on the assumption that the rising price level will not have any other effects on the economy; if wages do not keep up with inflation, volumes have to fall. Also, tax as a % of GDP will generally rise as nominal GDP increases.

      If I have any more comments, it makes sense to put them on your blog, and not this one.

      Delete
    7. Actually, there is a node to adjust the real GNP as the inflation rate goes up. I have it go down as that is what usually happens.

      Delete
  4. Hi Nick, congratulations with your blog, a real contribution as far as I can judge. Based on my limited knowledge (not beiing an economist) I believe I can agree with your ideas on the role of models, but with some remarks:

    - You state "as simple as possible whilst retaining the complexity to push my understanding a bit further". There is some contradiction in this statement as each simplification will bring a model further away from the (extremely complex) reality and thus make it less meaningfull.

    - You state "With simple theoretical models, we are trying to understand a very specific mechanic." However, such simple models may lack the emergent properties of more complex models, which may play a very important role in such mechanics.

    - You state "If we don't understand the results, they are useless". That is very true, but again, these results could be quite different from what you would expect due to emergent properties.

    I am curious about what you think about these points.

    Anton van de Haar

    ReplyDelete
    Replies
    1. That's a fair point. I think the problem is that we cannot get away from making assumptions and simplifications in economic models. In particular, most models will rely to some extent on assumptions about human behaviour. There are no hard rules we can rely on, so we just have to take a view on what people will do and use that. Some people try to take what they see as a scientific approach here, but in reality their assumptions are no less ad hoc than anything else.

      This is fine, but it's important that we understand how our conclusions depend on those assumptions. Otherwise we don't know how seriously to take our results. If the model has emergent properties, that's great if we can then understand those and decide if they apply in real life. But if we can't tell if they're really just down to our assumptions then we can't rely on them. So we have to go with what we can understand - anything else is too unreliable.

      I agree though that this is a problem if we consider that many things that happen are a result of complexity. In fact, I think that many important things about the way the economy behaves are overlooked because they involve differences between people and are therefore much harder to model. Finding ways to understand these things is really important.

      Delete