Stamp duties forecasting models based on intra-annual data

This paper addresses the issue of providing timely forecasts for disaggregated fiscal data. In the aftermath of the 2008-2009 crisis governments have become increasingly aware of the importance to generate trustworthy, time-consistent budget forecasts. Recently, several papers (Onorante et. al., 2008; Pedragal & Perez, 2010; Hallet et. al., 2012. Ghysels & Ozkan, 2012) showed that using intra-annual data can increase forecasting performance. In addition, these models can provide governments with "early warning signals". A second strand of literature (Lutkepo, 2010; Asimakopoulos et. al., 2013) recognizes that aggregating the forecasts for different expenditure and revenue components reduces the forecast error of the resulting budget deficit. Unfortunately, little evidence is available concerning the best models for specific revenue categories like stamp duties or inheritance taxes. In this paper we contribute to the literature by modeling the regional revenues of stamp duties using quarterly and monthly data covering the time period 1994-2013. More than 60 models were tested, including a-theoretical AR and (E(G)) ARCH models, and more theory driven DL, ADL, VAR, VEC models. In addition, combination forecasts were produced. Forecast horizons vary from one month ahead to 2 years ahead; model selection is based on the Aikake Information Criteria and on theoretical arguments, stemming from housing models. Performance evaluation is based on MPE, MAPE and RMSPE of both insample as well as out-of-sample forecasts.. In general, multivariate models show lower in-sample forecasting errors than univariate models. Based on the out-of-sample performance, theory driven models outperform the least complex models for shorter forecasting horizons. However, with forecasts heading for 2 years the outperformance of the more advanced models is less convincing. In addition, the results reveal that combination forecasts strongly enhance the accuracy of stamp duties revenues forecasting models. In general, the results are in line with the findings of Favero & Marcellino (2005) : simple forecasting models and combined forecasts outperform more complicated specifications.

from #MedicinebyAlexandrosSfakianakis via xlomafota13 on Inoreader http://ift.tt/1YqPUoP
via IFTTT