First Time: Cluster Analysis

http://www.crai.com/ecp/assets/cluster_analysis.pdf

This was the first article that formed my basic notions about cluster analysis and its applications to economic problems.

In this context, there was a need to conduct a problem-solution research. The immediate observation was that Time magazine had cut its cover price. There were two plausible reasons put forth to explain this price cut: it could be part of an anti-competitive predatory pricing strategy to hinder the entry of the Independent into its segment, or it could be a strategy to enter the mid-market segment occupied by the Daily Mail and Daily Express.

Of course, the price cut may be a calculated attempt by Time to achieve the above dual objectives, allowing it to defend its segment whilst gaining a foothold on another.

This proves a headache for competitive policy regulators. As a regulator which advocates competition, it is a boon for Time to enter the mid-market segment but a bane if Time’s actions are stifling competition in the upmarket segment. This presents a trade-off between policy objectives- should the regulator step in to order a reversal of the price cut, it will have eliminated potential competition in the mid-market segment. On the other hand, allowing the price cut may send the message that regulators will accommodate anticompetitive behaviour when a firm employs strategies that induces such policy dilemma, emboldening firms to carry out more anticompetitive practices in the future.

However there is a greater concern even before they consider the policy dilemma: Is there sufficient proof of anticompetitive behaviour by Time in the first place?

In an ideal situation, econometric analysis can be employed to examine the price elasticity and cross-elasticity of demand of Time and the Independent. If the cross-elasticity of demand for the Independent (in relation to the price set by Time) exceeds a certain benchmark, it will warrant further examination for potentially anticompetitive effects (even if that was not the initial motive of Time) arising from the price cut.

Reality, however, is far from ideal. The lack of historical price variability severely cripples the internal validity of the regression. The results obtained are unlikely to reflect the actual correlation between prices and demands in the real world.

Although Lexecon was writing this article from the perspective of a consultant (whose objective is to help the unnamed client to examine whether Time’s price cut imposes sufficient pressure on the client to warrant a legal complaint or a counterattack), it provides a glimpse of how Cluster analysis is conducted. 

Cluster analysis starts by defining consumer characteristics. The consumer characteristics are assigned to every product under examination. There will be data collected on the proportion of consumers who holds the characteristics defined for each product.

Initially, each product has its own cluster. The consumer data of product will be evaluated from a multi-dimensional software which utilizes an iterative computational approach to determine the “distance” between each product. The closest products will be combined into higher clusters iteratively.

The results of cluster analysis allows analysts to predict the switching behaviour under the assumptions that (i) Consumers of a current product tend to switch to rival products which targets similar key consumer characteristics. (ii) Ceteris paribus, not taking into account other plausible effects such as demographic shifts in preferences, variations in price sensitivity between consumers and other exogenous factors.

I hope to read more and refine my understanding in the weeks to come.