Per Ranebo och Hannes Lindbeck, presenterar sitt examensarbete

Predicting the aluminium spot price on the London Metal Exchange

Abstract

When a large wholesaler purchases metal, they may tie the contract
price to the spot price on the London Metal Exchange (LME). Hence it
is of great importance for the company to make accurate forecasts of
the LME spot price, and this Master's thesis aims at finding a model
that can improve the forecasting of the aluminum price on LME. The
work is done in cooperation with Bröderna Edstrand Group.

First the significance of a number of macro economic variables, such
as price of electricity, the UK interest rate and the UK GDP for the
spot price is investigated. Mathematical models of the metal price
time series in general with or without transformation of data and
nonstationary time series in particular are evaluated.

The results show that none of the proposed macro economical variables
could improved the perfomance of the spot price prediction. Instead
the preferred approach to handle the trends has been to use
deterministic trends sequentially detected with a
CUSUM-detector. Furthermore the deviations from the CUSUM trends show
dependence and conditional heteroscedasticity. The dependence is
handled with an AR-approach and the conditional heteroscedasticity is
handled with a GARCH-approach as part of the modelling.

The CUSUM detector gives a new and interesting approach to model
financial time series, since the model can be adjusted to detrend data
in different ways using trends with different characteristics.

The results show that modelling data with an AR - GARCH model works at
least as good when data has been detrended with a CUSUM detector as
when data has been transformed with a Box-Cox transformation.