Data-driven methods for equity similarity prediction

Many applications rely on the accurate prediction of company similarity to be effective. Diversification avoids similarity for risk reduction. Hedging through equity-neutral investing seeks similarity in order to ensure risk in long positions is effectively offset by short positions, and vice-versa. Relative valuation requires formation of a 'peer group' to which financial ratios, such as the P/E ratio, can be compared. This article considers two data-sets that have not traditionally been used for this purpose: sell-side equity analyst coverage and news article co-occurrences. Each is shown to have predictive power over future correlation, a key measure of future similarity. It is further shown that the analyst and news data can be combined with historical correlation to form groups that are on par or even exceed the quality of the Global Industry Classification System, a leading industry taxonomy.