Scottish housing market: tax revenue forecasting models – review

Application

forecasting

Good. There are examples of all three technical assumptions used to forecast housing prices and transaction volumes: rules of thumb, growth accounting models, or consensus forecasts.

policy

Fair. Univariate models forecast how prices and transactions typically move over time, but do not consider the underlying causes. Neither of the three technical assumptions excel at fiscal costings. There is some scope for adjusting assumptions for fiscal sensitivity tables or scenario analysis. Alternative capacity for fiscal impact costing would need to be developed.

Accuracy

short run

(quarters one to eight)

Fair. Evidence suggests consensus forecast averaging limits can produce better forecasts than any one model on its own. Rules of thumb and growth accounting are useful for unpredictable series; however, there is sufficient evidence that more sophisticated models are useful for forecasting housing markets.

medium run

(years three to five)

Fair. There is little difference between rules of thumb and growth accounting models for the short and medium run. Some evidence suggests that the consensus average forecast is less accurate than simple univariate approaches and regression models.

Communication

story telling

Fair. Technical assumptions vary in their ability to tell a story. Generally, they are not tied to economic theory in a way that permits forecast revisions and the path of the forecast to be well articulated. However, assumptions such as the growth rate of GDP or the consensus survey can capture general economic trends.

transparency

Good. All three approaches can be made independent and transparent (provided the consensus forecast is viewed as an externally-imposed rule. However, the underlying methodologies and assumptions used to produce the individual forecasts underlying the consensus would typically not be available.

Data compatibility

Fair. Rule of thumb and growth accounting frameworks work well with quarterly data and are generally well suited for medium-run forecasts. However, few Scotland-specific external forecasts of the housing market are produced on a quarterly basis for a five-year time horizon.

Resources

Good, with a qualification: internal capacity may still be required for policy analysis. However, for forecasting purposes, a technical assumption is well suited for a small tax as a share of the overall budgetary revenues. Easily estimated or imposed in spreadsheets and statistics software packages.

Evidence

Alquist and Vigfusson (2013), An de Meulen et al. (2011), Auterson (2014), Batchelor (2007), Belsky et al. (2007), Case and Shiller (1987), Crawford & Fratantoni (2003), Granziera et al. (2013), Makridakis (2008), McCue (2009), An de Meulen et al. (2011), Moro and Nuño (2012), Tsolacos (2006).

Application

forecasting

Good. Univariate models can be used for both prices and transactions. They can produce ex ante forecasts without needing to be conditioned on auxiliary forecasts of exogenous variables.

policy

Poor. Because they use their own history-with no links to other economic or fiscal variables-univariate models are not suitable for policy analysis by themselves. They may, however, be used as a component of a model that can perform policy analysis, such as forecasting the parameters of the housing price distribution to impose on the housing base.

Accuracy

Short run

(quarters one to eight)

Good. Evidence suggests that in the short-run, univariate time series models routinely beat other forecasting procedures and models with more structure.

Medium run

(years three to five)

Fair. Evidence and theory suggests that as the forecast horizon lengthens, univariate models are likely to be relatively less reliable than models grounded in theory and equilibrium concepts. The length of the useful forecast horizon is determined by the speed of decay, which is rarely greater than 8 quarters; however, their ability to capture simple trends beyond two years may prove sufficient for the housing market.

Communication

story telling

Poor. The models are mostly a black box without explanatory variables or theory to provide interpretable coefficients.

transparency

Fair. Models are estimated by applying relatively little judgment. Model equations can be published; however, they are usually updated frequently and coefficients and specification may change faster than documentation. Because the model structure is data driven and not intuitive, it could pose a challenge for budget scrutiny.

Data compatibility

Good. Univariate models work well with the number of observations of quarterly data available for the Scottish housing market.

Resources

Good. Univariate models require few resources and little expertise to estimate. Similar to current methodology and may not require more resources. Easily estimated in modern software packages such as Eviews and Stata. Cumbersome in Excel.

Evidence

Barari et al. (2014), Bollerslev (1986), Box and Jenkins (1976), Brooks and Tsolacos (1999), Brown et al. (1997), Crawford and Fratantoni (2003), Engle (1982), Enders (2014), Fullerton et al. (2001), Hyndman (2012), Lis (2013), Maitland-Smith and Brooks (1999), An de Meulen et al. (2011), Park and Hong (2012), Stock (2002), Stevenson and Young (2007), Tong (1983).

Application

forecasting

Fair. Potential to produce forecasts which closely match the true tax base. Many examples in the literature for prices and transactions. However, because they rely on exogenous variables, they cannot produce ex ante forecasts, reducing their score.

policy

Good. Can include a wide range of variables representing government policy and other explanatory variables that are well-suited for policy analysis, if they are specified with that in mind.

Accuracy

short run

(quarters one to eight)

Fair. Because they emphasize theory, they can miss important dynamics and useful information in the data that a theoretical specification for causal inference may miss. Can include lags of explanatory variables to improve short-run dynamics.

medium run

(years three to five)

Fair. Similar to short-run. Accuracy of medium run depends on accuracy of exogenous variables.

Communication

story telling

Good. Multivariate models are well-positioned to provide a narrative for the forecast path and forecast revisions. Explanatory variables are intuitive, parameter signs fit with economic theory, and causal interpretations are often possible.

transparency

Good. Multivariate models can be estimated with little judgment beyond which variables to include and the specification. Model equations can be published. Relationships are grounded in theory and unlikely to change. If they change it would be in magnitude of coefficients, not sign or overall specification. Model specification is intuitive and permits useful debate among budget stakeholders and fiscal watchdogs.

Data compatibility

Fair. Multivarite models work well with the number of observations of quarterly data available for the Scottish housing market. There may be some limitations on potential explanatory variables.

Resources

Fair. Few analysts with some specialised knowledge. A small increase in complexity compared to the current model process. May require more than current resources. Easily estimated in Excel and modern software packages such as Eviews and Stata.

Evidence

Auterson (2014), Bank of England (2008), Carnot et al. (2014), Case and Shiller (1987), Dicks (1990), Fullerton et al. (2001), Makriditis et al. (2008), Meen (2013, 2012, 2009, 2001, 1990), Poterba (1984), Tsolacos (2006).

Application

forecasting

Good. Examples have been found in the literature for both prices and transactions. They can produce ex ante forecasts without needing to be conditioned on auxiliary forecasts of exogenous variables.

policy

Poor. VARs are constructed under the principle of parsimony-the lowest number of variables that are significant. If a variable is not useful, then it should not be included. VARs generally lack causal economic and fiscal links to do policy analysis. Some shocks (impulse response) analysis possible.

Accuracy

short run

(quarters one to eight)

Good. Under ideal circumstances, VARs would offer superior forecasting potential, particularly for short-run forecasts. This is conditional on having sufficient Scottish historical data that doesn't limit the VAR's specification.

medium run

(years three to five)

Fair. Evidence and theory suggests that as the forecast horizon lengthens, VAR models are likely to be relatively less reliable than models grounded in theory and equilibrium concepts. The length of the useful forecast horizon is determined by the speed of decay.

Communication

story telling

Poor. The models are mostly a black box without a causal specification grounded in theory. Often have counterintuitive signs and the long lag structure makes interpretation difficult.

transparency

Fair. Models are estimated mechanically by applying relatively little judgment. Model equations can be published; however, they are usually updated frequently and coefficients and specification may change faster than documentation. Coefficients tend not to be intuitive and could pose a challenge for budget scrutiny.

Data compatibility

Fair. Possible to estimate, depending on number of explanatory variables and lags included, but small sample size makes difficult to estimate a model. Is suitable for quarterly forecasting if seasonal dummies are included or deseasonalised data is used. Can produce 5-year forecast, but performance is better for short-run.

Resources

Good. VAR models require few resources and modest expertise to estimate (with modern software). Similar to current methodology and may not require more resources. Easily estimated in software packages such as Eviews and Stata. Cumbersome in Excel.

Evidence

Bernanke et al, (2004), Brooks and Tsolacos (2010), Sargent and Sims (1977), Stock and Watson (2002).

Application

forecasting

Good. Examples have been found in the literature for both prices and transactions. They can produce limited ex ante forecasts depending on specification.

policy

Fair. Not ideal for policy analysis, but depending on specification and integration in wider macroeconomic model can do impulse responses and limited policy analysis.

Accuracy

short run

(quarters one to eight)

Fair. ECMs are generally regarded as offering good short-run forecasting potential.

medium run

(years three to five)

Good. Because they are based on equilibrium concepts, ECMs are generally regarded as more capable at five-year horizons than other approaches.

Communication

story telling

Good. ECMs are generally interpreted easily, signs and magnitudes are meaningful, and a consistent narrative can be formed.

transparency

Fair. Equations can be published and their specification (particularly model coefficients) is unlikely to change frequently. Model equations are more opaque to budget scrutinizers with only a general economics background than more simple regression equations, but the equations are nonetheless more economically intuitive than VARs.

Data compatibility

Fair. Possible to estimate, but low sample size for parameter confidence, especially given the crisis. The current approach is likely more appropriate for the sample size. Can handle quarterly frequency data.

Resources

Fair. Can be estimated and maintained with relatively few analysts. But specialized knowledge and advanced degree in economics or statistics recommended.

Evidence

Addison-Smyth et al. (2008), Auterson (2014), Enders (2014), Riddel (2000, 2004), Stevenson and Young (2014)

Application

forecasting

Fair. Generally, not well-specified for best forecasts or correspondence to the true tax base. But can be complemented by a richer suite model framework to arrive at a rich specification for both prices and transactions.

policy

Fair. Not suited for fiscal impact costing. Can assess fiscal multipliers, responses of government intervention, dynamic scoring. Often on a different accounting basis (national accounts rather than commercial accounting standards).

Accuracy

short run

(quarters one to eight)

Fair. Near-term is based on monitoring; however, because national accounts data is subject to revisions, the starting point and recent history may not be pinned down. Rely on dynamics more than the multivariate econometric models of Subsection 3.3, but generally still grounded in theory.

medium run

(years three to five)

Fair. Because of its theoretical underpinnings and extensive reliance on long-run equilibrium conditions (closing of the output gap) and frequently levels variables (with heavy reliance on error correction models), these models are likely to improve upon naïve forecasts for the medium run. However, medium run forecast will suffer from the same concerns as the short run.

Communication

story telling

Good. Can produce consistent, intuitive narratives in-line with economic theory of the housing market and its impact on the macroeconomy. Coefficient signs make sense.

transparency

Fair. Estimated with considerable judgment, especially for smoothing the near term (first two quarters) and squaring with economic monitoring. Model equations can be published. Relationships are grounded in theory and unlikely to change. If they change it would be in magnitude of coefficients, not sign or overall specification. Model specification is intuitive and permits useful debate among budget stakeholders and fiscal watchdogs.

Data compatibility

Fair. The Quarterly National Accounts of Scotland should provide enough information for a simple model of GDP; however, the specification of the housing sector (residential investment) and accompanying aggregate behavioural equations is likely to be rudimentary.

Resources

poor. One or two very experienced and skilled analysts is sufficient to maintain a large-scale macroeconometric model once developed; however, the development is a significant undertaking. Useful to have several less experienced analysts in charge of economic monitoring and to work independently on model components such as import/export modules. Would require programmable statistics software beyond Excel.

Evidence

Auterson (2014), Gervais and Gosselin (2014), Carnot et al. (2014), Case et al. (2005), Robidoux and Wong (1998), De Vroey (2016), Granger and Newbold (1986), Hicks (1939), HM Treasury (2016), Elming and Erlmler (2016), Klein and Goldberger (1955), Mankiw and Weil (1989), Manuelli and Sargent (1988), Meen (2012), OBR (2013), PBO (2016), and Romer (2016).

Application

forecasting

Poor. While some versions are suited to forecasting, few examples have been found in the literature for housing prices and transactions. Very unlikely could be specified well enough to forecast the LBTT tax base. Can perform ex ante forecasts: conditional variables jointly determined within the system.

policy

Fair. DSGE models may have use for modelling the transmission through the economy of housing market scenarios. DSGE models are not appropriate for static policy costing (costings that are estimated using a single market and do not consider the feedback effects of the rest of the economy). However, they can be used for dynamic scoring (modelling the impulse of government revenue changes through the wider economy).

Accuracy

short run

(quarters one to eight)

Fair. Evidence is mixed and depends on focus (small models generally outperform larger models) and specification.

medium run

(years three to five)

Fair. Some evidence to suggest that DSGE models may offer better performance in the medium run than other models less grounded in economic theory

Communication

story telling

Fair. Theoretical foundations lend themselves to intuitive narratives. However, complexity of interactions runs significant risks of becoming a 'black box' with difficult or no interpretation.

transparency

Poor. Results and assumptions can be opaque and harder to explain to policymakers. Considerable judgment is applied throughout estimation. External scrutiny would require specialist training.

Data compatibility

Good. A DSGE model could be estimated using the Quarterly National Accounts of Scotland. Small DSGE models can be constructed on as few as seven data series; however, incorporating the household sector would require the addition of two or three others. All are available.

Resources

Poor. Although be more compact and take less effort than a large-scale macroeconomic model. Most institutions with DSGE models maintain a full suite of other models to complement and compare with the DSGE's results, as it is generally too aggregated to provide the detail required for budgeting and policy. It still requires forecasting exogenous variables, monitoring for the current period and near short-run.

Evidence

An and Schorfheide (2007), Burgess et al. (2016), Caldara, Harrison and Lipinska (2012), Carnot et al. (2011), Christiano et al. (2005), De Vroey (2016), Del Negro and Schorfheide (2012), Herbst and Schorfheide (2015), Iacoviello and Neri (2010), Kydland and Prescott (1982), Romer (2016), Sargent (1987), Slanicay (2014), Slobodyan and Wouters (2012), Smets and Wouters (2003), Smets and Wouters (2007).

Application

forecasting

Poor. Microsimulation models cannot in themselves produce a forecast for the housing market. They require a model to forecast the baseline assumptions (growth factors) for variables that determine the evolution of receipts. However, they are useful for forecasting the distribution of the prices over time, given these inputs.

policy

Good. Microsimulation models are ideal for assessing the impact of different tax rates and thresholds on government revenues. A drawback is that they are by nature mechanical and require ad hoc adjustments to the simulation output to incorporate behavioural responses.

Accuracy

short run

(quarters one to eight)

N/A. As the model itself cannot forecast, this criterion is not applicable. However, if economic variables forecast with auxiliary models are accurate, then the microsimulation model should give accurate conditional estimates of revenues.

medium run

(years three to five)

N/A. ibid

Communication

story telling

Good. Because microsimulation models are a rote imposition of the tax code on real households, communication of microsimulation results are intuitive.

transparency

Good. The underlying equations are mechanical identities, and aside from weights to scale results to the population level, little estimation and no judgment is applied.

Data compatibility

Poor. There are limitations for forecasters' access to tax-payer level data in Scotland. Microsimulation model development may not be possible currently, but could be considered in the future if alternative data protocols are explored.

Resources

Poor. Microsimulation models are likely to require considerable resources to develop and maintain.

Evidence

Best and Klevin (2013), Edwards, (2004), O'Donoghue (2013), OBR (2016), Morand et al. (2010), Sauerbier (2002), Wolfson and Rowe (1998).