Krill-Herd Support Vector Regression and heterogeneous autoregressive leverage: evidence from forecasting and trading commodities

Abstract

In this study, a Krill-Herd Support Vector Regression (KH-vSVR) model is introduced. The Krill Herd (KH) algorithm is a novel metaheuristic optimization technique inspired by the behaviour of krill herds. The KH optimizes the SVR parameters by balancing the search between local and global optima. The proposed model is applied to the task of forecasting and trading three commodity exchange traded funds on a daily basis over the period 2012–2014. The inputs of the KH-vSVR models are selected through the model confidence set from a large pool of linear predictors. The KH-vSVR’s statistical and trading performance is benchmarked against traditionally adjusted SVR structures and the best linear predictor. In addition to a simple strategy, a time-varying leverage trading strategy is applied based on heterogeneous autoregressive volatility estimations. It is shown that the KH-vSVR outperforms its counterparts in terms of statistical accuracy and trading efficiency, while the leverage strategy is found to be successful.

Keywords:

• Krill Herd
• Support vector regression
• Commodities
• ETF
• Leverage

Notes

¹ The maximum induced speed of equation (15) and the foraging speed of equation (17) are set to 0.01 and 0.02 ms⁻¹, respectively, as Gandomi and Alavi (2012) suggest.

² Intraday traders, speculators and hedge funds are examples of short-term agents. Commercial banks can be though as medium-term agents, while pension funds and insurance companies are long-term agents.

³ Interest costs are calculated by considering a 0.56% interest rate p.a. (the Euribor rate at the time of calculation) divided by 252 trading days. In reality, leverage costs are also applied during non-trading days so that we should calculate the interest costs using 360 days per year. But for the sake of simplicity, we use the approximation of 252 trading days to spread the leverage costs of non-trading days equally over the trading days. This approximation prevents us from keeping track of how many non-trading days we hold a position.