The forecaster as a storyteller

Paul Krugman explaining the IS-LM model:

[T]he first thing you need to know is that there are multiple correct ways of explaining IS-LM. That’s because it’s a model of several interacting markets, and you can enter from multiple directions, any one of which is a valid starting point.

Which beautifully illustrates this point:

There seem to be two ways of understanding things; either by way of a metaphor or by way of a story, …[and] the metaphorical and narrative explanations answer to each other.

The metaphors McCloskey refers to are what we commonly term ‘models’; the narrative is the story that justifies the model’s existence. The metaphor provides a framework for the narrative, as the narrative provides a context for the metaphor. The importance of this interdependence is that models are empty without a narrative to explain why they exist. Equally, a historical narrative is of little help if it doesn’t give rise to a generalisable metaphor that we can use to simulate counterfactual worlds.

Krugman’s introduction to IS-LM (a macroeconomic metaphor) illustrates not only the importance of the narrative but also the fact that multiple narratives can support the same metaphor. This ties in really well to Matt’s discussions of the value of economic forecasting. The value in forecasting is not in the predictions that our models make about the future, but in the usefulness of the narrative the forecaster tells. That narrative varies across forecasters depending upon the relative importance of various actors in their story, despite the fact that they all have pretty much the same model of the economy in mind. That may help explain why they all tell a different story and yet profess to largely agree about all the important issues.

  • And the advantage of using the economic method to do this is that the assumptions made are transparent 🙂

    I’m not sure why you thought I would disagree with this at all – given it is exactly the type of argument I’ve long held onto for economic modeling. 

    • I think the difference between your descriptions and the rhetorical approach here is the separation of model and story. I thought you might consider the story a consequence of the model, which might result in one uniquely ‘right’ story.

      • There are a few things that prevent that from being the case:

        1)  The fact that models are generally made to answer specific questions – there is no general model

        2)  The fact that there is missing data, implying that even if everyone was using the same objective model and data there is room for subjectivity in tying things together.

        3)  The model informs how you look at the data just as much as the data informs what model to pick.

        Without a general model and a perfect information set, I find the idea of a uniquely “right” story disingenuous. 

        However, for a well specified question, and a given set of defined exogenous variables, we may be able to tie down a uniquely right story – but to me this relies a lot on how we define the question 😉

         

        • Sure, but doesn’t only (2) matter for forecasters at the margin? They’re all trying to answer the same questions using pretty much the same data and models, after all.

        • I wouldn’t say that most forecasters are trying to answer the same question per see.  Even if we were to say this was the case with “macro” forecasters, the difference is usual due to differing views of exogenous variables – where the dearth of relevant data is endemic and the appropriate way to model them is relatively disagreed upon 😉

          In this context I would also note that, contrary to the way we often here macroeconomists argue they do actually provide relatively similar forecasts – I would say that the marginal value a forecaster adds has more to do with their ability to identify what elements will be more important for their client, and help to translate these forecasts into a usable and identifiable way to deal with risk.  This makes the narrative component essential not just for helping to create the model – but for helping to create the output that clients value, which in turn explains why forecasters “sound” so different when their numbers are actually quite similar.