Download files
Abstract
The productivity slowdown across industrialised countries since 2004 is a persistent puzzle. This thesis proposes new productivity decompositions, using both industry and firm-level administrative data sets, with a focus on innovative techniques to advance understanding of the drivers of firm, industry, and country productivity performance. Particular attention is paid to micro-theoretic foundations of the proposed techniques, and to the rigorous application of appropriate econometric and data science techniques.
At the industry level, drivers of productivity change are identified from a micro-theoretic framework implemented using an index number approach. Drawing on a non-parametric model, Chapter 2 decomposes productivity growth into explanatory factors for 12 selected industries and 16 market sector industries in Australia. Technical progress is found to support increasing productivity, though its contribution is partly offset by production inefficiency. Production inefficiency is interpreted as lagged output, inactive operation or possible measurement errors on a case-by-case basis. The overall performance of productivity growth and its explanatory factors is affected by the market shares based on a weighted average industry aggregation.
In addition to the industry-level productivity decomposition in Chapter 2, a firm-level productivity decomposition is developed in Chapter 3 for a market that consists of incumbents, entrants and exiters. This new method enables decompositions of productivity into components to be merged with firm dynamics. The framework is applied to Australian firm-level data and reveals the dominant contribution of incumbent firms to industry productivity and industry efficiency. A difference-in-differences approach is proposed that validates the firm dynamics from the counterfactual perspective.
Price imputation is essential when detailed price information is unavailable to support productivity decompositions. Chapter 4 introduces tree-based machine learning models for estimating missing prices in cases where there is product entry and exit, or product “churn”. Model performance metrics from (electronic-point-of-sale) scanner data confirm the prediction accuracy of tree-based models. An economic explanation is proposed to link the decision tree structure and consumer preferences. Tree-based models are recommended for price imputation due to their prediction accuracy and compatibility with consumer utility types.