The dynamic nature of many asset price processes and
the lack of perfect hedging assets can lead to unstable hedge ratios over time,
ne- cessitating the re-estimation and rebalancing of cross-hedges. Cross- hedging
occurs when a portfolio or asset is hedged with a statistically related yet not identical underlying derivative.
Ordinary Least Squares regression is an oft applied method for estimating
constant minimum- variance hedge ratios to curb price volatility or manage a
market-neutral porfolio. However, constant estimates are often unsuitable under
cross- hedging where the dependence structure between the two assets change over
time. Rather than traditional correlation-based hedging, this paper focuses on cointegration-based cross-hedging with
respect to the equilibrium between asset prices. We apply and test the
out-of-sample efficacy of models that enable the cointegrating vector, or hedge
ratio between two nonstationary price series, to vary over time. Models are
estimated across daily data for selected equity, bond and commodity pairs.
Rolling- window regression, exponentially-weighted moving
average and Dynamic Linear Models (Gaussian Linear State-Space Models) are
investigated. Results show that time-varying parameter models have superior
out-of- sample hedging performance compared to constant parameter methods. This
finding is confirmed through extensive
Monte Carlo simulation. In practice, this reduction in basis risk comes with
incurred transaction costs from routine hedge rebalancing.