Almost ten years after writing my master's thesis on quasi-hyperbolic utility in dynamic general equilibrium models, my view of the theoretical framework I used has changed considerably.
At the time, I was interested in introducing more realistic intertemporal behaviour into standard macroeconomic models. Quasi-hyperbolic discounting seemed attractive because it allowed present bias and time inconsistency to enter models that otherwise assumed highly rational behaviour.
Over the following years, however, I became increasingly dissatisfied with neoclassical modelling. Part of this dissatisfaction came from its empirical performance, and part from the level of abstraction required to make the models tractable.
The representative consumer and the representative firm are obvious examples. They are useful simplifications, but they compress enormous differences in behaviour, expectations and circumstances into a single theoretical agent.
The concept that troubled me most, however, was the utility function.
I increasingly found it difficult to accept an approach in which behaviour is represented as if individuals continuously maximized a smooth mathematical object such as
\[ U = U(c). \]
I understood the standard argument that utility does not have to represent happiness or pleasure and can simply be interpreted as a numerical representation of preferences. But over time this explanation became less convincing to me.
I have always been more attracted to approaches that begin with observable behaviour rather than with hypothetical internal constructs. From this perspective, utility often appeared to me as a convenient theoretical device whose form was selected mainly because it allowed the mathematics to work.
Recently, however, I came across the work of Wolfram Schultz on reward and decision signals. This made the concept somewhat more interesting to me again, because it suggested that quantities resembling subjective value and utility may be reconstructed from observed responses rather than simply assumed in advance.
That distinction matters to me. I remain sceptical of beginning with a chosen utility function and deriving behaviour from it. I find it much more appealing to begin with behaviour and ask what kind of valuation structure can be inferred from it.
I have also become increasingly uncomfortable with the pervasive use of derivatives in economics. My objection is not mathematical. Calculus is extraordinarily useful. The problem is interpretative.
Human behaviour is usually observed through finite actions and transitions: buying or not buying, accepting or rejecting, saving more or less, waiting or acting. The infinitesimal change that appears naturally in calculus is therefore a mathematical construction rather than something we literally observe in social behaviour.
For this reason, I would now be more cautious about treating continuity and differentiability as natural properties of economic behaviour. I would rather see them as approximations that may be extremely useful in some models, but that should not automatically be confused with the underlying phenomenon.
My dissatisfaction with the closed-economy assumption has changed even less.
Closed economies are analytically convenient, but almost no modern economy is truly closed. Goods, capital, labour, information and financial claims continuously cross borders. Domestic saving and investment are connected to the external sector, and capital accumulation cannot always be understood as a purely domestic process.
The same concern applies to models without money. I understand perfectly why economists begin with simplified real models. But money, credit, financial claims and payment systems are not secondary decorations added to an otherwise complete barter economy. They are part of the institutional structure through which modern economies actually function.
Despite these criticisms, I have never completely abandoned the type of models I studied in my master's thesis.
Their greatest strength remains the explicit treatment of time. Dynamic models force us to think about sequences, delayed consequences, accumulation and the relationship between present actions and future outcomes.
This remains extremely valuable.
What I have become sceptical of is not dynamic modelling itself, but the tendency to associate it with a particular package of assumptions: representative agents, smooth utility functions, continuous optimization, closed economies and highly simplified institutional settings.
If I rewrote my master's thesis today, I would therefore keep the intertemporal perspective but change the methodological starting point.
I would begin with observed behaviour, heterogeneity and actual economic institutions. I would treat utility as something to be estimated or reconstructed rather than simply imposed. I would use continuous mathematics where it provides a useful approximation, but I would avoid treating infinitesimal variation as an intrinsic property of human behaviour.
And I would be reluctant to describe an economy as closed or moneyless unless the simplification was essential to the specific question being studied.
Ten years ago, I was mainly interested in modifying the utility function inside an existing theoretical framework.
Today, I am more interested in asking whether the framework itself starts from the right objects.
That is probably the biggest change in how I now think about economic modelling.