Scottish housing market: tax revenue forecasting models – review

1. Context and scope

This report was commissioned by the Scottish Government in response to recommendations by the Scottish Fiscal Commission to explore model options for forecasting the housing market for Land and Buildings Transaction Tax ( LBTT). ^[1] The research provides a comparative evidence base for forecasters to decide whether to change the current approach, and how to proceed should they decide to do so. ^[2] The review's specification has been jointly agreed by the Scottish Government and the Commission.

We focus on the residential tax base (that is, average prices and transactions volumes). Techniques for assessing the housing price distribution and applying tax rates to the base are addressed to a lesser extent. That said, the models and extensions we review can be applied to the wider economic and fiscal forecasting framework.

Budget forecasting is important for identifying trends that could affect the ability of the government to deliver its policy goals both in the immediate future and over the longer term. For this reason, it is important that forecasts be accurate. But the requirements of public sector forecasting models go beyond accuracy. They should also lend themselves to intuitive and transparent communication of revisions to stakeholders. Model developers must also be aware of practical constraints such as the availability of reliable and timely data, and should choose models that require appropriate resources to develop, run, and maintain. We attempt to capture these considerations in our evaluation. Our methods and criteria are described in Section 2.

Section 3 provides formal model assessments for eight general classes of models that have an evidence base of application to the housing market or may offer modelling potential in the future. These are: 1) technical assumptions, 2) univariate time series approaches, 3) multivariate econometric models, 4) vector autoregressive models, 5) error-correction models, 6) large-scale macroeconometric models, 7) dynamic stochastic general equilibrium models, and 8) microsimulation models.

Section 4 describes models and techniques that could be integrated alongside the assessed model classes in the budget forecasting framework to improve forecasts. These include Bayesian techniques, dynamic factor modelling, computable general equilibrium models, techniques to anticipate turning points (booms and busts), and forecasting the tax take directly.

Models for public budgeting come with trade-offs that are not only difficult to balance, but often at opposition to one another. For example, models that perform well at predicting the path of future revenues may not be designed to accurately estimate the causal relationships between variables and may underperform for fiscal impact costings and risk assessments. To deal with these competing objectives, many practitioners in budget institutions maintain a suite of different housing models: one or more for predicting tax revenue and social expenditure, one or more for policy analysis (fiscal impact estimates and distributional analysis), and one or more for producing the macroeconomic forecast. The Scottish forecasting framework may similarly be best served by more than one approach. Section 5 describes how practitioners use these models together in forecasting frameworks.

The practical issues unique to public sector forecasting can heavily influence model selection and are not always given weight in academic forecasting literature. For this reason, we also conducted a series of practitioner interviews on the practical matters of public forecasting. Responses are provided in Section 6.

Section 7 summarises the results of the review and briefly discusses the next steps, should development of a new approach be taken forward.