Gold, oil, and stocks

In this paper we analyze number of aspects in the dynamics of the prices of gold, oil and stocks over 26 years from January 2, 1987 until December 31, 2012. Analysis is performed on both intra-day and daily data. We employ a variety of methodologies to study dynamic co-movements among assets (realized volatility, DCC GARCH, vawelets), lead relationships (Granger causality methodology), and long-term equilibrium relationships (Johansen cointegration methodology). In terms of of the dynamic analysis we employ a novel time-frequency approach: we analyze co-movements not only from a time series perspective but also from the investment horizon perspective simultaneously. Hence, we are able to bring unique evidence on how correlations among major assets vary over time and different investment horizons.

Specifically we show that correlations between assets are low or even negative at first, they later increase, and change in their patterns becomes most pronounced after decissive structural breaks take place. The change hints at the existence of the decissive structural breaks in the correlation structure between the pairs of assets that are endogenously found in late 2000’s. Further we show that correlations before crisis exhibit differences in pattrens at different investment horizons but during and after crisis they exhibit large swings and differences in correlations at shorter and longer investment horizons become negligible. The result indicates that there is no room for risk diversification in the sense of various investment horizons after 2008.

Granger causality methodology was employed to analyze lead trelationships among assests. The overall finding is that in all three pairs we observe mainly feedback causality and no strong one-way Granger causality is present consistently for entire period of our sample. Finally, using a Granger causality approach we assess how fast markets process information available in prices of assessts. We show that untill 2001 markets process information immediately as the contemporaneous effects are strong. After 2001 it takes up to three days for markets to process available information.

We analyzed long-term equlibrium relationship among assests using Johansem methodology while accounting for structural breaks. Endogenously dectected break in a correlation series for specific pair of assets is taken as a break to divide sample of data to test for cointegration between the same pair of assets. For all periods and all pairs of assets the result is unique: no cointegration has ben found. The rusult corroborates our earlier finding of low correlation among assets.