Moving to more sophisticated application of quant methods, there is another distinction between long-term and short-term quants. To best explain this difference, let's first look at some similarities. Typically, both will build portfolios based on combining multiple signals they believe offer an expectation of outperformance over time. Unlike blind quants, both tend to focus sharply on portfolio construction and trading costs to reduce the noise around their intended signals, and both go through a process of continuous improvement looking for new signals that will be additive. Each struggles with the inherent danger of data mining and of unearthing false signals that look good based on past performance but make zero or even negative contributions to future returns.
Long-term quant managers, however, seek signals that last usually six months or longer and will outperform over a market cycle. Their signals generally have some intelligible underlying rationale that supports their conviction for application throughout a cycle. Indeed, much of what they target are the same attributes a fundamental manager considers in the analysis of an individual stock. The difference is a fundamental manager (whose holdings are generally much more concentrated) must also evaluate what is idiosyncratic for each stock, whereas a long-term quant uses diversification to minimize idiosyncratic risk and maximize exposure to their targeted signals. Some of these managers (our firm being one of the pioneers in this area) may shift the balance between different signals dynamically in different market conditions to further enhance return.
In contrast, short-term quant managers seek patterns that will work now, even if their shelf-life is unpredictable. Some may be as short as weeks or just days, as others latch onto the same trends. In many cases, these short-term signals rely on artificial intelligence involving complex pattern recognition. As more and more players chase the same prize, it can be assumed that the average signal life will shorten even more as it gets arbitraged away more quickly. In turn, ever increasing computer power is required with ever diminishing payoff. Big data is especially appealing in this arms race, but the risks of data mining are exponentially higher because many short-term signals have no fundamental driver. Some signals may have no predictive power at all, or may erode and then invert if the manager cannot jump off the moving train fast enough. Although the very best may continue to excel, most likely there will be many more losers than winners.
Of course, we have had to simplify a bit in the service of our taxonomy. In practice, individual managers may offer a combination of long- and short-term approaches. For example, a quant with roots intrinsically in the long-term camp (like our firm) may utilize big data techniques to identify new signals, or to aid with risk management or the timing of trades. Yet, to reduce the risk of data mining, there will still be a requirement for an abiding rationale behind any new signal uncovered and that it stand up to out-of-sample analysis.
Whether or not this version forms the right basis, we would urge all our fellow quants to join us in our desire to develop a more sophisticated nomenclature to help our clients better understand our differences and avoid the disappointments that have befallen other parts of our industry.