New York – In the world of quantitative modeling, there are a number of solid approaches that have been developed over the years. Quantitative research firms employ mutifactor models to determine value, EVA models, or employ econometric techniques (to name a few) to structure models that can separate good companies from bad. The scores assigned by the rating system can then be used to rank the individual categories.
Most quants use a decile ranking system, which means that there are ten buckets a company can reside within. Equity research performance evaluation systems tend to convert these deciles into Buy, Hold and Sell categories for the purposes of estimating recommendation accuracy. We note that this approach has several drawbacks, which we cover below.
First, the analysis needs to be differentiated according to the sector, or indeed the characteristics of the balance sheets of companies. This point is one that the quants have generally covered.
Second, the classification system may assess that a firm, for example, is weak, but this does not reflect any judgment as to the timing of a price decline. Indeed, this is precisely why more successful multifactor models include a technical factor/s, such as the momentum factor in the Carhart multifactor model, to time the sell recommendations. Once the technical timing metric/s are added the profitability of the models improves sharply.
Third, the translation of the decile classification system into Buy, Sell, Hold recommendations may be appropriate over a long time frame (consistent with the backtesting period), but may not be consistent with the current market environment. Typically, the Buy recommendations are those companies that are in the top 3 deciles, the Sells are the companies that are in the bottom 3 deciles and the rest are considered to be Holds.
In periods where major trends are affecting the market, however, the entire decile structure could be a Sell or a Buy. This indicates a need for descriminant analysis. Discriminant analysis seeks to find the most relevant distinction between data sets, for example in credit analysis, trying to find the dividing line between companies that are more likely to fail than survive over a specific timeline than others in the data set. Superimposed on the decile structure, descriminat analysis could move the B/S/H lines so that a down market might have a declie structure as follows: from 0% to 70% would be sells; 70% to 85% could be Holds; and only the top 15% of the firms under coverage may be considered Buys.
Despite the above issues, quant models have a good track record in providing consistent returns over time (i.e. the returns tend to be less volatile than many other research approaches). What we have found is, that for many quant research firms, if the top 20 to 50 stocks and the worst 20 to 50 stocks are used to create a portfolio much better returns can be achieved.