AIS data and the price of oil : a study of predictive feasability

2018

Compared to the oil market, physical movement of oil-carrying vessels is very precise and

reflects the real production rates more timely than official reports. In this paper, we examine

whether detailed information on crude oil movements, obtained from AIS tracking system,

can be used to better predict the oil price. We use a variety of model specifications and

introduce a novel instrument for the role of expectations to this question. This instrument is

based on vessel speed, and it offers insights into the apparent lack of empirical indications of

speed optimization. We show that the AIS data can contribute to predicting the oil price. We

also explore Kilian’s (2009) hypothesis that the model of the oil price should include three

factors: expectations of future prices in addition to supply and demand. We triangulate his

instrument with one that we construct independently.

We have explored several different specifications for the relationship between inter- and

intraregional oil ship traffic and the oil price and found that a statistically significant

relationship exists. Our findings indicate that the correlation between variables make OLS an

unsuitable tool for this analysis since endogeneity bias will suppress the actual relationship.

However, we have found that the relationship is robust to different VAR specifications. The

contribution to explanatory power as measured by Factor Error Variance Decomposition is

marginal, but it might still be a small improvement on present methods. We have examined

the apparent paradox of non-optimal ship speed behavior. We found that a less stringent

specification apparently resolved the paradox; the freight rates do indeed influence ship speed

if lags and correlation are allowed. The short time-span is preventing us from conclusively

saying that the issue is resolved, but it appears at least to be worthy of further investigation.

We assessed the validity of using ship opportunity cost as a measure for GDP. While we cannot

address the question of possible bias created by Kilian’s use of the Baltic index, we

nonetheless offer conceptual support, as our unrelated instrument for the same opportunity

cost showed strong statistical significance.