A Generalised Model of Investment under Uncertainty: Aggregation and Estimation

We propose a structural model of investment which is based on the aggregation of (S,s) investment projects within firms. This encompasses the findings that whilst firm level investment is smooth, plant level investment is lumpy and frequently zero. We undertake stochastic aggregation and derive a structural firm level investment estimator. The empirical performance and fit of this estimator on a panel of manufacturing firms is encouraging and provides an avenue for general policy simulation. This model also explains the rich non-linear dynamics of firm level investment data and the frequent simultaneity of firm level investment and disinvestment. This approach provides an alternative structural estimator to the standard convex adjustment cost models, such as Tobin's Q and the Euler equation. The is important because these estimators, which assume quadratic adjustment costs, appear to be misspecified and subject to a fallacy of composition between smooth firm level investment and lumpy plant level investment. For completeness we also consider time aggregation as an alternative source of smoothing but statistically reject this as being insufficient to smooth investment alone. This test also rejects most plant level data, such as the US\ LRD and UK\ ARD, as being generated from a single (S,s) process.