Twenty-two years of Japanese institutional forecasts

Abstract

This paper evaluates accuracy and rationality of real GDP forecasts made by 38 Japanese private institutions over the past 22 years. It finds that about 80% of current-year forecasts and year-ahead forecasts made by them pass various tests for rationality. Moreover, the encompassing test reveals the following results: (a) All of these forecasts outperform the naïve forecasts; (b) About half of their current-year forecasts are inferior to the corresponding forecast of VAR, VECM, or the Japanese government; (c) Almost all of their year-ahead forecasts are significantly superior to the corresponding forecast of VAR or the Japanese government, but one-third of them are significantly inferior to VECM forecast; (d) The consensus forecast outperforms typical institution's forecast.

Notes

 Batchelor and Dua (1991), Jansen and Kishan (1996), Joutz and Stekler (2000), and Ashiya (2002c) detect some bias. On the other hand, Zarnowitz (1985), Artis (1996), Pons (1999, 2000), Oller and Barot (2000), and Kreinin (2000) do not. See also Fildes and Stekler (2002) and references therein for the related literature.

 Ashiya (2002a, b; 2003) also uses the data from ‘Monthly Statistics (Tokei Geppo)’.

 See Keane and Runkle (1990) and Stark and Croushore (2002). We obtain the same results by using the revised data of g_t released in June of year t + 2.

 Ashiya (2002c) also uses these forecasts as a benchmark.

 Average forecast error of typical forecaster is about 0.5 percentage points for current-year forecast (

) and 1.3 percentage points for year-ahead forecast (

). See Ashiya (2002d) for the details.

 Ashiya (2002d) finds that the consensus (i.e. mean) forecasts (

and

) are not biased.

 Ashiya (2002d) finds that the consensus forecasts (

and

) use available information efficiently.

 See Chong and Hendry (1986), Mizon and Richard (1986), Ericsson (1992), and Jansen and Kishan (1996).

 When δ < 0 (δ > 1), f ¹(f ²) is so inaccurate that the minimum squared error composite requires ‘short selling’ of it (Cooper and Nelson, 1975, p. 8).

 The generalized encompassing test

was also conducted, and similar results obtained. See Ashiya (2002d) for the details.

 Ashiya (2002d) finds that the consensus forecast for the current year (

) encompasses all of naïve forecasts, VAR forecast, VECM forecast, and the government forecast.

 Ashiya (2002d) finds that the consensus forecast for the next year (

) encompasses naïve forecasts, VAR forecast, and the government forecast, but that it does not encompass VECM forecast.