由於美國近期受到次級房貸風暴的衝擊,本研究首先採用傳統線性因果測試,探討房地產價格、利率與股價三者中兩兩變數之間的因果關係。期間始於西元2003年1月至2008年12月止,代理變數為:美元3個月期倫敦銀行間拆款利率、美國20大都會區房屋價格指數與美國S&P500股價指數。經由傳統方法發現:利率與房地產價格短期存在雙向回饋因果關係;而股價受到利率與房地產價格之影響,僅存在單向因果關係,即房地產價格先影響利率走勢,而利率之變動再影響股價。另外亦將資料期間分段,並分別探討市場穩定期(前期)與次貸風暴期(後期)二者之間的異同與比較,並發現受次貸風暴事件影響於後期三變數之間存在有較明顯之因果關係。由此可證實在整個次貸風暴並逐漸演變成金融風暴期間,確實存在著結構轉變的情況。 而本研究第二部份,首先利用Enders and Siklos (2001)門檻共整測試以確定兩兩變數之間是否存在共整合(長期均衡)關係;同時並應用門檻自我迴歸模型 (TAR)和動能門檻自我迴歸模型(M-TAR)進行測試,以期找出最適模型。最後再利用門檻誤差修正模型(TECM)和動能門檻誤差修正模型(M-TECM)進行實證分析,以期發現兩兩變數於門檻上下區間是否存在著不對稱調整之非線性因果關係。研究結果發現利率對於房地產價格變動所造成的衝擊,存在不對稱行為反應,此由於美國房地產市場的崩塌造成利率政策(利率走勢)的不穩定,因此利率對於房地產價格,存在不對稱因果關係。同樣地,房地產價格最後間接影響並導致2008年股價(股票市場)的崩盤,股價對於利率變動所造成的衝擊,亦存在不對稱非線性的衝擊反應特性。
This paper aims to investigate the causal relationships among real estate prices, interest rates and stock prices in the U.S. during the period of Subprime Mortgage Crisis. In addition, the relationships among the changes in these three variables considered are analyzed. The empirical results show that the three proxy variables – 3 month Libor interest rates (LIBOR), S&P/Case-Shiller Home Price Index (SPCS20), and N.Y. S&P 500 Stock Index (S&P500) do not share a long-run common trend during the period from Jan. 2003 to Dec. 2008. However, for the short-run effect, a bi-directional causality was found in the preceding relationship between SPCS20 and LIBOR. Among other causal relations, it is found that LIBOR and SPCS20 “Granger-cause” S&P500, but not in reverse order. This confirms that changes in real estate prices lead the changes in monetary policy (interest rates), and interest rates lead stock prices. The reduced interest rate effect in 2007-2008 periods seems to have been negated by anticipated collapse from housing market. The interest rates together with real estate prices play a leading role to the movements of stock prices during the period of subprime mortgage crisis. Furthermore, the generalized impulse response function and variance decomposition are adopted and implications of the findings are discussed. Next, in consideration of the structure change may result in different causal relations before and after the Subprime Mortgage Crisis, the whole sample period is then separated into two sub-periods. Cointegration relationships have been found for both sub-periods (2003/1 ~2006/2 and 2006/3 ~2008/12). Besides, there exist different causal relationships compared for the first and second sub-periods. This confirms the structural change assumption during the crisis period. The nonlinear techniques are further used in this research. Firstly, from the applications of the threshold autoregressive models--TART and M-TART models, respectively, the results show that there exist threshold cointegration relationships between LIBOR and SPCS20, also between S&P500 and LIBOR. Secondly, based on the results from the threshold error correction or momentum-threshold error correction models (TECM or M-TECM), the null hypotheses of no cointegration are rejected for the three pair-wise cases, suggesting that there has been a force of recovering equilibrium or a co-movement through time for each pair-wise case. Likewise, the findings that there exist asymmetric causal relationships between LIBOR and SPCS20, also between S&P500 and LIBOR, are consistent with the previous empirical evidences from the TART and M-TART models. Furthermore, the significant dynamic adjustments of LIBOR to SPCS20 shocks show markedly different responses above and below the threshold value. This indicates the speed of adjustment towards long-run equilibrium is much faster in the higher regime than in the lower regime. The asymmetric adjustment to equilibrium argues that LIBOR responds asymmetrically to SPCS20 shocks, but not in reverse order. For the relationship between S&P500 and LIBOR, based on the M-TECM, they clearly illustrate bidirectional short-run and long-run causal relationships exist between these two variables. Moreover, the significant asymmetric adjustment in S&P500 is consistent with our previous finding from the M-TART model, which argues that S&P500 respond asymmetrically to LIBOR shocks. At last, only a unidirectional causal relationship running from SPCS20 to S&P500 in the long run can be found between these two variables. However, there is no significant asymmetric causal relationship between S&P500 and SPCS20.