Causality in demand: a co-integrated demand system for trout in Germany

Abstract

This article focuses on causality in demand. A methodology where causality is imposed and tested within an empirical co-integrated demand model, not prespecified, is suggested. The methodology allows different causality of different products within the same demand system. The methodology is applied to fish demand. On the German market for farmed trout and substitutes, it is found that supply sources, i.e. aquaculture and fishery, are not the only determinant of causality. Storing, tightness of management and aggregation level of integrated markets might also be important. The methodological implication is that more explicit focus on causality in demand analyses provides improved information. The results suggest that frozen trout forms part of a large European whitefish market, where prices of fresh trout are formed on a relatively separate market. Redfish is a substitute on both markets. The policy implication is that increased production of trout causes a downward pressure on fresh trout prices, but frozen trout prices remain relatively unaffected.

Acknowledgement

The authors would like to acknowledge financial support from the Nordic Council of Ministers to the project ‘Price formation of freshwater fish species’.

Notes

¹ A data series is stationary if it moves randomly around a constant mean over time (i.e. mean and variance are independent of time) and is nonstationary if the value of the present observation depends on the value of former observations.

² Demand systems with well-defined preference structures include the Almost Ideal Demand system (Deaton and Muellbauer, 1980a) and the Rotterdam system (Barten and Bettendorf, 1989).

³ In the present article, theoretical consistency refers to own price effects being negative and of reasonable size.

⁴ A stationary data series is said to be integrated of degree zero (i.e. I(0)). A nonstationary data series is said to be integrated of degree one (i.e. I(1)) if its first differences (the difference between two periods) move randomly around a constant mean over time and integrated of a higher order (i.e. I(z)) where z ≥ 2, if the value of the present observation depends on the value of former observations.

⁵ In the case that the system is just identified restrictions can be imposed, but no testing is involved.

⁶ Several and probably most articles in the literature on co-integration use univariate tests like the Augmented Dickey–Fuller (ADF) tests to test for presence of unit roots, thereby determining order of integration. Typically, by pre-testing all variables individually in levels and differences. According to Juselius (2006), however, such tests are inappropriate to identify order of integration. Multivariate tests must be used. Therefore, in the present article both stationarity and the presence of I(2) are tested in a multivariate framework. The hypotheses of the multivariate tests for stationarity are ‘reversed’ in relation to the ADF test, since the null hypothesis is the presence of stationarity. The null is accepted if p > 0.05.

⁷ The price of redfish would have been 70% * (7.5% − 0.6%) = 4.8% higher, corresponding to a quantity of trout which was 1.41% * 4.8% = 6.8% higher.