금융변수의 불황예측력 비교 (Comparing the Predictive Power of Financial Variables on Recessions)

서지정보

· 발행기관 : 한국금융학회
· 수록지 정보 : 금융연구 / 27권 / 1호 / 29 ~ 69페이지
· 저자명 : 이근영

초록

본 연구에서는 간단한 probit 모형을 이용해 장․단기 금리격차, 주가, 환율 등 국내외 금융변수들이경기불황을 어느 정도 예측할 수 있는가를 살펴보았다. 실증분석결과 전반적으로 장․단기금리격차가 실물변수는 물론 주가나 환율변수보다 불황예측에 우수하다.
표본내 추정의 경우종속변수와 설명변수간의 시차가 커질수록 지속적으로 MSE가 커지고 수축기와 확장기의 불황확률평균차가 작아진다. 반면 표본외 성과가 우수한 1변수 모형의 경우에는 1개월 또는 12개월 앞보다3개월 또는 6개월 앞 불황예측시 MSE가 더 작고 불황확률 평균차가 더 커진다. 특히 MSE와불황확률 평균차 검정, 그리고 방향예측검정 등을 통해 예측결과를 종합적으로 비교해 본 결과6개월 전 장단기 금리스프레드인 국고채(3년)-CP유통수익률(91일)과 미국채(10년)-콜금리(1일)를같이 사용하는 경우 표본외 예측력이 가장 우수한 것으로 나타났다.

영어초록

This study investigates how well domestic and foreign financial variables such as term spreads, stock prices, and foreign exchange rates can predict domestic recessions by using simple probit models. It first uses in-sample estimation and out-of-sample forecasting methods in comparing mean square errors(MSEs) of these predictions. It also tests whether mean differences of recession probabilities between expansion and contraction periods are statistically significant or not. Finally, the Henriksson and Merton test is carried out to measure directional accuracy.
According to the in-sample empirical results, as time lags between dependent and explanatory variables are increasing, mean square errors are persistently bigger and mean differences of recession probabilities are decreasing. In addition, when mean square errors are smallest, mean differences of recession probabilities are biggest.
In case of the one variable models with better out-of-sample predictive performance, mean square errors are smaller and mean differences of recession probabilities are bigger in forecasting three or six months ahead recessions rather than one or twelve months ahead recessions. On the other hand, without relation to time lags, the two variable models with good out-of-sample predictive performance have bigger mean differences of recession probabilities between expansion and contraction periods than the one variable models. The two variable models also have bigger mean differences of recession probabilities in forecasting three or six months ahead recessions rather than one or twelve months ahead recessions.
The spread of short-term interest rates over the yield on a long-term bond generally outperforms stock prices and foreign exchange rates as well as real variables in predicted recession probabilities. Stock prices are useful in short-term ahead prediction, while spreads between domestic short-term interest rates and the U.S. long-term Treasury yields are better in longer term ahead prediction. Probit models forecasting recessions that use both stock prices and term spreads also have better out-of-sample predictive performances than models with the term spread alone.
The forecasting results based on comparison of mean square errors,tests for mean differences of predicted recession probabilities, and the Henriksson and Merton test suggest that the probit model using both the spread of commercial paper rate over the yield on a three-year Treasury bond(tb3_cp) and the spread of call rate over the yield on a ten-year U.
S. Treasury bond(ustb10_ca) has the best predictive power at the six months ahead horizon, compared to the other models and horizons.

참고자료

· 없음