Credit Crisis and the Price of Gold: Evidence from a Forecast Based Modified Granger Causality Model

Abstract

In this paper, based on the existing literature on gold in terms of its interaction with other financial assets as an asset by itself to diversify a traditional portfolio and also as a hedging instrument against adverse market movements and crashes, a proposal is made for an investigation into the causal dynamics between gold price and the movements of the prices of other financial assets within the premise of a new interpretation of the ECM based Granger Causality model with a simple rearrangement of terms entering the ECM and the interpretation of the results from such a system within the premise of theory of inferred causation and a graphical methodology associated with the same. Two versions of the model based on rearrangements and substitutions based on the long run equilibrium suggested by the cointegration regression between them with gold as the dependent variable to render an a priori bias to the model are run within a constrained VAR in levels setup by employing the SUR methodology to estimate the same. From the results of the first model that the author proposes to call the causal model, a TSCM (Time Series Causal Model) graph is constructed which is based on the theory of inferred causation and the related DAG methodology for a graphical interpretation of the same. A similar graph that the author calls the TSFM (Time Series Forecast Model) graph is constructed based on the results of the second VAR in levels model that the author calls the pure forecast model. Ex-ante out-of-sample forecasts for this pure forecast model are also presented with the forecasts spanning the period before the start of the credit crisis up until the January of 2011. The tests are repeated with a different choice for the in-sample time period spanning the core crisis period and ex-ante forecasts are made similar to the previous run of the model. The TSCM and the TSFM graphs for both the time periods are then employed to make interpretation of the causal links between the variables within the model. Based on these results the paper finds a causal link running from 7-10 year US treasury yields and the 20 year US treasury yields to gold price movements throughout the period of investigation. The paper also finds that though the causal links are more complicated for the period spanning the core of the crisis period, the causal flow from US treasury yields to gold price movements has been consistent throughout the period in light of the quality of the forecasts made by the pure forecast model for gold for the out-of-sample period for the two sets of tests conducted.