本研究建立了一個新的分析架構,藉由預測誤差來認定預測機構在「景氣循環」波動中的兩種可能「行為偏誤」現象-意圖性的樂觀或悲觀的預測態度而造成的行為偏誤與非意圖性地對新進訊息反應不及或反應過度而造成的行為偏誤,並透過此架構分析台灣經濟成長率的預測表現。我們的實證結果顯示,台灣的官方機構(行政院主計處)除了對景氣循環波動中的訊息反應不及而產生預測誤差外,在景氣擴張時期還會因過度悲觀而低估經濟成長率;而民間機構(中華經濟研究院)的預測則因為滲入更多意圖性的行為偏誤而導致其預測值較不準確,其於擴張期低估成長率的原因主要來自於過度悲觀及對訊息反應過度,而於衰退期高估的原因則來自過度樂觀和反應不及兩種行為偏誤。除此之外,我們還發現,行政院主計處的預測在2001年前後期有相當大的轉變,2001年之後預測行為偏誤現象更為顯著且伴隨較大的預測誤差波動。最後,我們建立相對應的「行為偏誤修正模型」,修正所認定出行政院主計處的行為偏誤並進而提出修正後的經濟成長率預測值。其中,修正了景氣循環過程中每一階段可能產生的行為偏誤。修正後的預測值明顯優於行政院主計處的預測值及文獻上利用大量訊息而建立的更新預測值。由於我們的分析架構僅需預測機構的預測值序列即可操作,因此我們認為,本文所提出的分析架構除了在相關研究上有所貢獻外,應可作為研究者於進行經濟預測時的另一項新選擇。
In this study, we propose a new approach to investigating forecasters' behavioral biases during business cycles via inspecting and analyzing one-step-ahead forecasting errors. For Taiwan's economic growth rate, the effects of optimism/pessimism and overreaction/under-reaction to the news on forecasts during the expansion and contraction periods of real activity are identified in the proposed framework. Overall, our empirical results discover that Taiwan's government usually under-reacts to the news during business cycles, and tends to be pessimistic, especially in expansions. Besides, these behavioral biases in the post-2001 period are much more significant than those in the pre-2001 period. On the other hand, when comparing the government to the private sector in Taiwan, these behavioral biases affect the private sector's forecasts much more severely. Moreover, in our in-sample and out-of-sample experiments, the forecast performance improves significantly after correcting the identified behavioral biases of the Taiwan government's forecast. Because the proposed approach needs only time series data of forecasts, it is easy to implement, and thus might be a good alternative to the existing approaches of economic forecasts.