In this era of ‘big data’, it is not the data itself that is transforming how we live, but the algorithms employed to collect, make sense of and exploit this data. Many sophisticated investment managers have learnt to use algorithms as a powerful tool in their investment armoury, and for quantitative (quant) managers, algorithms are a powerful differentiator of their style and a cornerstone of their investment approach.
The power and limitations of algorithms
An algorithm is a set of step-by-step instructions, written in computer coding language, to use a set of inputs (data) to produce one or more outputs. Given data about investment markets and a specific investment idea, an investment manager can design an algorithm to filter and sort the specific data inputs relevant to the idea, perform the appropriate calculations using the data and produce a set of portfolio constituents, trade list or other output reflecting the investment idea. For example, a manager attracted to large-cap Australian equities with price momentum may purchase the S&P/ASX 200 constituents as a time series data set and write an algorithm that calculates the price momentum of the different stocks, creating a suggested holdings list of higher-momentum stocks, comparing this list to the current stocks and weights in the portfolio and generating a trade list to move to the new target portfolio. While all this could be done manually (without algorithms), it is easy to see how the investment process becomes faster and less prone to error when algorithms do the hard work. Using algorithms instead of investment professionals can also reduce costs and (sometimes) aid transparency.
The thought discipline that coding up an investment idea requires is also very useful – one must be very precise in articulating the idea (in an ‘objective function’), the variables that impact it and the degree of influence (‘weighting coefficient’) of each variable. The algorithm also helps the investment manager adhere to the manager’s investment philosophy and does not allow fear, greed or any other human sentiment into the equation. For large fund investors who increasingly want a custom rather than ‘one size fits all’ solution, it is relatively easy to adapt the algorithms to reflect the personal preferences of the investor (although not all managers are willing to do it). For example, for a superannuation fund who likes the abovementioned momentum strategy but not the added concentration risk, the manager could introduce a step into the algorithm which more equally weights the desired stocks. Or, by adding a risk model to the S&P/ASX data set (inputs), the manager could limit the algorithm’s potential outputs to portfolios which do not breach upper bounds in relation to particular portfolio risks, say, by capping higher momentum stocks in the financial sector at a 20% overweight of the benchmark weight to financials.
Three investment problems observed in the market
However, the dazzling power of algorithms should not blind us to their limitations. The ‘flash crash’ of May 2010, where the Dow Jones Industrial Average crashed 1,000 points in a few minutes, is thought to have been caused by High Frequency Trading algorithms which competed to drive trade bids lower and lower because … well, they did what they were programmed to do.
1. Performance dispersion between similar quantitative strategies
Part of the reason may be that the algorithms that drive many strategies are fitted to the back tests that worked, not the ones that didn’t. For example, in the new breed of systematic alternative risk premia harvesting (global macro) strategies, one of the fundamental factor risks de-emphasised in many live strategies is the ‘Value’ factor, notwithstanding its academic support and intuitive appeal, simply because of its recent poor performance. Algorithms can be the contagion that transmits research data-mining (the selective use of data to fit an idea) into a live portfolio.
Managers of strategies need a sensible narrative about what is, and what is not, in the algorithms that drive their portfolios.
2. Unintended bets driving a portfolio’s outcomes
Algorithms can be used to specify a ‘maximise return’ objective function. These algorithms can seem elegantly simple but equity strategies which use them to tilt to specific factor bets like Value, Growth, Size and Yield often suffer from a large amount of their portfolio risk not coming from these intended bets, but other unintended bets. The algorithms do exactly what they are coded to do. There is no risk awareness if risk inputs are not included in the algorithm. If an investor is surprised to find their income-tilted portfolio loaded with Australian financials or US utilities, a diversification risk-oblivious algorithm may be to blame.
3. Outperformance dragged down by complexity and transaction costs
The power of algorithms is such that sometimes a manager wants every aspect of the investment idea captured in and exploited by the algorithm (just one more data input, one more variable …). This can lead to a strategy inundated with trade signals. It is hard to find an algorithm adept in balancing the expected performance contribution of a signal against the transaction costs of trading. It is even rarer to find an algorithm that accurately calculates the tax costs of trading and balances these against the merits of trading on each and every signal.
Due diligence on managers who use algorithms should not ‘stop at the door’. Any manager who won’t answer reasonable questions about their algorithms on the basis that it is a proprietary ‘black box’ is out of step with need of large fiduciary investors to address opaqueness in their investment portfolios. Transparency equips fiduciaries to better understand and justify the risks of the investment, whether the rewards are sufficient compensation and whether the fees are sensible.
Our experience is that investment professionals (the human kind!) are key partners to the coders who build algorithms for investment portfolios. They can check the reliability of the data inputs and how secure the software and systems are. They can provide a wise reality check over the output produced by algorithms before the investment recommendations hit the market.
Raewyn Williams is Managing Director of Research at Parametric Australia, a US-based investment advisor. This is general information only and does not consider the circumstances of any investor. Additional information is available at parametricportfolio.com.au.