Revenue Forecasting Practices: Differences across Countries and Consequences for Forecasting Performance

Transcription

1 Revenue Forecasting Practices: Differences across Countries and Consequences for Forecasting Performance Thiess Buettner and Bjoern Kauder CESifo, Munich Summer Symposium 2010 Centre for Business Taxation

2 Revenue Forecasts Key role for budgeting and policy Recent experience, once more emphasizes macroeconomic fluctuations: US-CBO: 2008 revenues 7.5% below 2007 forecast US-OMB: 2008 revenues 5.5% below 2007 forecast Ireland: 2008 revenues 15.4% below 2007 forecast Beyond macroeconomics: institutional characteristics?

3 Determinants of Revenue Forecasting Performance US states revenue forecasts Feenberg, Gentry, Gilroy and Rosen (1989) Boylan (2008) Bretschneider, Gorr, Grizzle, and Klay (1989) Krause, Lewis, Douglas (2006) Literature on EU Stability and Growth Pact, mainly concerned with deficit forecasts (Strauch, Hallerberg, von Hagen, 2004, Jonung and Larch, 2006, Pina and Vedes, 2007) Literatur on developing countries (e.g., Danninger, 2005) emphasizes weak institutions.

4 Outline This paper: performance among selected OECD countries Institutions matter: Timing Tax structure Independence from possible gov t manipulation

5 Outline This paper: performance among selected OECD countries Institutions matter: Timing Tax structure Independence from possible gov t manipulation Forecasting performance Forecasting conditions Institutions and independence Empirical analysis

6 Performance Measurement Revenue forecasts aims at informing about revenue constraint Performance usually assessed using forecast errors E i (F i X i ) Decomposition into different sources not available

7 Performance Measurement Revenue forecasts aims at informing about revenue constraint Performance usually assessed using forecast errors E i (F i X i ) Decomposition into different sources not available Mean forecast error: MNFE 1 N N i=1 E i

8 Performance Measurement Revenue forecasts aims at informing about revenue constraint Performance usually assessed using forecast errors E i (F i X i ) Decomposition into different sources not available Mean forecast error: MNFE 1 N N i=1 E i Standard deviation: SDFE 1 N N (E i MNFE) 2 i=1

9 Performance Measurement Revenue forecasts aims at informing about revenue constraint Performance usually assessed using forecast errors E i (F i X i ) Decomposition into different sources not available Mean forecast error: MNFE 1 N N i=1 E i Standard deviation: SDFE 1 N N (E i MNFE) 2 i=1 Root mean squared forecast error: RMSFE SDFE 2 + MNFE 2

10 Descriptive Statistics on Forecast Errors Country MNFE SDFE RMSFE Obs. (Fiscal-) Years Austria Belgium Canada /98-09/10 France Germany Ireland Italy Japan /98-08/09 Netherlands , New Zealand /98-07/08 United Kingdom /98-07/08 USA: CBO /97-07/08 USA: OMB /97-07/08 Average

11 Forecast Errors by Year

12 Forecast Errors by Country/Institution

13 Determinants of Forecasting Performance Some more obvious candidates Macroeconomic uncertainty: GDP forecast error Timing Tax structure: segmentation into different taxes Types of taxes Changes in tax law/administration Methodology: macroeconomic models, microsimulations

14 Forecasting Conditions Country Time Span Taxes for GDP Forecast error (Median) 50% rev. MNFE SDFE RMSFE Austria Belgium Canada France Germany Ireland Italy Japan Netherlands New Zealand United Kingdom US: CBO 8.5 US: OMB Average

15 Institutions and Independence Assignment executive branch of gov t legislative branch independent bodies / research institutes Inclusion of independent experts government representatives Macroeconomic forecast political parameter done by forecasters Other aspects: competing forecasts, EU budgetary surveillance

16 Institutional Characteristics and Independence Research Ext./Gov. Macroecon. Indepen- Country institutes experts forecast dence Netherlands Austria Germany USA: OMB USA: CBO Canada New Zealand Belgium United Kingdom Japan France Italy Ireland

17 Determinants of Forecast Error (1) (2) (3) (4) (5) (6) Constant (.517) (.463) (4.02) (3.62) GDP forecast error (.316) (.323) (.322) Time span (.167) (.150) log No. of taxes for 50% of revenue (1.12) (1.01) Independence (2.29) (2.06) Italy (1.88) (1.67) Canada (1.73) (1.53) R Observations Dependent variable: One-year-ahead forecast error for total tax revenues. Insign. country dummies suppressed.

18 Determinants of Precision, Total Revenues (1) (2) (3) (4) (5) (6) Constant (1.78) (7.47) (1.67) (6.38) (1.56) (1.81) Time span (.190) (.210) (.171) (.192) (.167) (.177) SDFE for GDP (1.20) (1.69) (1.03) (1.44) (.984) (1.09) log No of taxes... (1.54) (1.32) Independence a (2.00) (2.01) (2.00) (1.94) R R 2 adjusted Observations Dependent variable is the standard deviation of one-year-ahead forecast error (SDFE) for total tax revenues. Weighted least squares estimates.

19 Determinants of Precision and Accuracy, Total Revenues Dependent variable SDFE RMSFE (7) (8) (9) (10) Constant (1.68) (6.22) (1.53) (5.72) Time span (.152) (.175) (.130) (.151) log No. of taxes for % of revenue (1.39) (1.24) Independence a (1.72) (1.84) (1.47) (1.55) EU-SGP (1.03) (1.27) (.873) (1.07) SDFE for GDP (.937) (1.32) RMSFE for GDP (.878) (1.30) R R 2 adjusted Observations

20 Determinants of Precision, Disaggregated Revenues Dependent variable SDFE (1) (2) (3) (4) SDFE for GDP Tax type 1 (Income taxes) Tax type 2 (Corporation taxes) Tax type 3 (Sales taxes) Tax type 4 (Others) Independence Time span Time span Tax type Time span Tax type Time span Tax type log No R Observations

21 Determinants of Precision and Accuracy, Disagg.Rev. Dependent variable SDFE RMSFE (5) (6) (7) (8) Tax type 1 (Income taxes) Tax type 2 (Corporation taxes) Tax type 3 (Sales taxes) Tax type 4 (Others) SDFE for GDP RMSFE for GDP Independence EU-SGP Time span Time span Tax type Time span Tax type Time span Tax type log No. of R Observations

22 Conclusions Mean revenue forecast error tends to be small generally no sustained biases Conditional on GDP forecast error pessimistic forecasts for Canada and Italy Substantial variation in precision of forecast errors More than 90% of differences in forecasting precision is attributable to Timing Precision of GDP forecast Tax structure (corporation and personal income taxes) Independence from possible gov t manipulation Caveat: policy changes not taken into account; tax reforms of income taxes might explain lower precision

23 Future Research Theoretical analysis of gov t manipulation Emphasizing time varying conditions Preliminary analysis points at electoral cycles

24 Forecast Errors and Elections (1) (2) (3) (4) (5) (6) (7) ECY (.011) (.012) (.010) (.010) (.017) ECY Ind..020 (.034) ENY (.011) (.012) (.010) (.010) (.016) ENY Ind..073 (.036) R Obs ECY: elections current year, ENY: elections next year. (1) to (3) include time and country/institution specific effects. (4) to (7) allow these effects to be tax specific.