Back to Portfolio for the Future

Active Risk Budgeting Gets Consistent Alpha

A new paper takes an experimental look at “Active Risk Budgeting,” a method of portfolio construction that looks to build upon older and sometimes passive risk budgeting approaches, adding enough active management to allow the risk budget to change over time. For example, an institution might want its risk budget to change to incorporate forecasts of expected Sharpe ratios or tactical tilts.

Some History

Risk-based indexing (RB) has been around for years. One of the classic early accounts of RB in scholarly literature, back in 2010, was the work of Paul Demey, Sebastien Malliard, and Thierry Roncalli, and was simply titled “Risk-Based Indexation.”

RB, that paper observed, breaks both with conventional market capitalization weighting (CW) and with fundamental indexing (indexing that weights the components of an index by such economic metrics as earnings or dividends.) RB, rather, weights to diversify the risk of a portfolio. This has its own variants: RB can include equally weighted portfolios or their minimum variance cousins.

Another paper worth mentioning in this connection, from 2013, was “The Smart Beta Indexing Puzzle.” The lead author of this one was Zelia Cazalet, a quantitative researcher at Lyxor Asset Management, Paris, and a lecturer at the Ecole Nationale de la Statistique et de l’Analyse de l’Information. Cazalet et al. set out the trade-offs involved in using RB rather than a CW portfolio. Cazalet observed that RB portfolios are by construction less liquid than CW. Furthermore, they have risks related to the passivity in passive management: tracking difference risk and tracking error risk. Still, RB portfolios are more diversified and less versatile than the more conventional alternatives, and the investors who have it under consideration will have to make that trade-off.

An Accurate Signal

Now, in 2019, a new paper by Gaurav Chakravorty as the head of a team at the investment management firm Qplum reports on “a series of systematic experiments [that] gradually increase the predictive accuracy of the input signal.” That is, this series of experiments (backtesting simulations) increases the extent to which the signal used for portfolio construction knows the Sharpe ratio of the next five days. The experiment led the authors of the paper to conclude that only at the limit, where the signal knows the Sharpe ratio perfectly, does the performance of the tangency portfolio [the portfolio recommended by the CAPM] catch up with the performance of Active Risk Budgeting.

Given such results, Chakravorty et al. recommend the use of Active Risk Budgeting in portfolio management, especially for active management strategies on derivatives.

Their Experiment

Their backtesting experiment involved the following conventions: they treated the allocations as determined using closing prices with transaction costs of three basis points per trade throughout. They rolled over the expiries to make sure that they would always trade the highest volume expiry for each contract. Also, they assumed that each portfolio construction method (CAPM and active RB) is run daily to obtain the desired allocation with daily rebalancing, with a scaling of leverage to target a constant 10% annualized volatility each day.

They used six futures contracts: e-mini S&P 500 futures; 10-year T-note futures; crude oil futures; copper futures, gold futures, and soybean meal futures.

Of course, only the first of these involves exposure to the equity markets, the second the sovereign debt. The third through the six sample a range of derivatives.

They conclude that “in a forward-looking low signal-to-noise environment,” CAPM, also referenced here as the “Max Sharpe strategy,” is likely to severely underperform what they recommend. Active RB is a “robust alternative to Max Sharpe strategy.”

The Authors

Gaurav Chakravorty was the co-founder of the CIO office at Qplum. As it happens, Chakravorty recently (in May if 2019) left Qplum to take on the post of software engineering manager, machine learning aspects, Google Assistant, at Google.

The othe authors are Ankit Awasthi and Mansi Singhal, both also of Qplum. Singhal, too, left her position there recently for a new position at BNY Mellon, senior principal, digital business development.