I am not a keen supporter of numerical modelling of economic phenomena. I have seldom seen key issues of controversy in economics resolved by numerical modelling and think that, as a policy tool, numerical modelling does not improve on sensible thinking through of the issues using low order non-numerical and even purely conceptual models. It is absolutely essential to try to understand things initially using a minimialist conceptual model. Many of the extremely abstract theoretical models that I studied as a graduate student to try to understand the world – e.g. infinite time horizon, representative agent, consumption-savings models – are now being fleshed out numerically and used in important parts of, in particular, macroeconomics.
Partly this reflects the need for research activity by academics and doctoral students who find it difficult to get established given the disappearance of low hanging (theoretical) fruit. Often the empirical implementations involve essentially arbritrary simulations or ‘calibrations’. Partly too the prelediction for numerical modelling comes from the unconstrained abilities of modern data processing equipment to run simulations – though not to think problems through – and through the provision of large scale data bases which offer the possibility of N! possible studies and through the preferences of investigators who find it easier to implement a software package than to think a problem through. A final reason for interest in these procedures is that public servants want numbers – however arbitrary – not insight.
A less essential – though to me compelling – ingredient in my dislike of heavy-handed numerical modelling is the extreme boredom I experience in listening to the discussion of numerical outputs in research seminars when I don’t understand how they were obtained. For years I have refused outright to attend a seminar on time series econometrics although on a few occasions I have been trapped by a misleading title. I’d prefer to dine on trap-door spiders or drink cheap chardonnay.
A difficulty in economics is that the character of an economic problem often depends critically on a difficult-to-determine elasticity. Building a numerically-oriented model which incorporates a host of economic relationships with numerous behavioural relationships conceals but does not alter this crucial dependence. Often the crucial parameter is something like a simple demand or supply elasticity. The investigators have not got a clue about this parameter and assume its significance will for some reason vanish when its role is imbedded in a large scale model with several dozen other less important, although unknown, parameters.
One disastrous area of empirical implementation of models has been finance – The Economist has an excellent discussion - where models developed to understand options markets (e.g. the Black-Scholes model) have been used to numerically value options. These models ignore at their peril the possibility of catastrophic events such as the collapse of the US mortgage market and their use changes the character of markets they are intended to describe. These models have not only not successfully described behaviour but have helped to inflict significant economic damage.
Dumbledore gets Snape to give Harry Occulumency lessons rather than doing it himself. Jack Economics
I had to scratch around to interpret this comment. and still mdon’t. Harry Potter fans anywhere?
Two thoughts. (1) it’s just spam (2) This is obtuse comment on things people can do. Dumbledore can’t do it, Snape can.
I think you’re dead wrong here Harry. For a start, how on earth can you do any sort of investment appraisal without crunching the numbers? And crunching those sort of numbers is numeric modelling of economic phenomena.
As I’ve often told people, the alternative to evidence-based formal quantitative modelling is not no modelling, but prejudice-based informal qualitative modelling. The human condition means we are forced to make predictions about the future all the time, those predictions are always made with some sort of mental model, so lets get the unexamined assumptions examined and the untested logic tested. Which is all that formal modelling does.
DD, Maybe I have been too strong in my remarks – if you have good enough data then go for it. But typically you do not – economic phenomena can’t be described with the precision of the physical sciences. I think that forcing yourself to do a cost-benefit analysis forces you to consider the magnitude of effects – sometimes jiggling the numbers shows that some effects will be important and others not so. But I seldom believe the results of numerical modelling – in my experience there are too many assumptions built-in that strain credulity. A major issue is forecasting demand for the output being provided by a new investment project. I prefer to have a general understanding of the situation ratherr than relying on numbers pulled out of the air.
there is a difference between what they do in calibration in economics and what they do in econometrics.
i love reading serious econometrics. things like estimation structural IO models or even macro time series. some of the stuff being done in australia is superb.
non-statistical calibration is typically incoherent.
I’d concede enough, Harry, to agree that the point estimates fallng out of models are often not what is most useful – the difference between point estimates of a base case and a counterfactual is most often what you’re after, in which case fixed errors can wash out. And also the real-life use of a lot of quantitative modelling is precisely the sensitivity testing – “jiggling the numbers [to show] that some effects will be important and others not so”.
But then again, I doubt BHP-Billiton depends on just “a general understanding of the situation” in choosing between putting a few billion into a new iron ore mine or a new offshore gas field.
Not sure Dd whether attaching numbers to the feasibility of a new iron ore project will help resolve the ‘go ahead’ decision. My guess is that qualitative insights – continued growth in China and scientific info re extraction costs will matter a lot.
Harry makes a great point highlighting the fact that numerical modelling makes a theoretical framework less general and subject to parametric uncertainty. In addition, it is also very difficult to listen to presentations pertaining to such analysis because it is difficult to run through all the specific assumptions behind these models. However, without such applications, the practical relevance of theory would be fairly minimal. In my knowledge, I would not know that many policymakers who would rely on optimal control output to formulate regulations. Moreover, numerical simulations are often required because many theoretical models as they are extended to provide a better description of reality become unwieldy. Hence, the “low hanging theoretical fruit” are not a primary motivator, rather it is the need for pragmatism among economists. I think Harry’s comments are very relevant if you are an academic economist trying to publish, but if you want to influence the real world, then numerics will be an important and pertinent part of your toolbox.