Algorithmic Allocation Risk Adjusted Return

Family offices often require clarity when adding an algorithmic allocation to their portfolios. They may also have questions regarding the parameters that distinguish different algorithms when choosing a strategy appropriate for their portfolio. This article summarizes the challenges to bear in mind when allocating capital to algorithmic strategies. It is intended for capital allocators tasked with managing multi-generational family wealth. If you are considering an allocation to an algorithmic fund or designing an in-house strategy, it is essential to think through various aspects, including:

What type of algorithmic model you will be using.

How correlated the strategy is with other investments in your portfolio.

The impact of taxes and fees.

How you will measure success.

Simulating returns on the strategy, and

Risk-adjusted and total returns.

Algorithmic Allocation Model

An algorithmic allocation model uses modern computer algorithms to manage investments over a range of timeframes, instruments, and strategies. The model seeks to improve the risk-adjusted return of the investment portfolio by enhancing the performance of an asset allocation portfolio, such as the 60:40 stock:bond portfolios that are popular with family offices today. In addition to seeking diversification amongst holdings within an asset class and allocations to asset classes, an algorithmic allocation could also partition risk amongst the categories of computational models available for managing the holdings.

Sometimes it is useful to visualize the relationship between individual holdings, asset class allocations, and algorithmic allocation as concentric layers of an onion. Each layer builds upon the prior, and each exists for a specific purpose. Holding diversification aims to reduce the business risk of any one investment failing and having an outsized impact on the portfolio; in the case of asset allocation, the goal is to reduce market risk by allocating capital to multiple asset classes in differing markets. However, with algorithmic allocation, risk management is equipped to handle both business and market risk. In fact, it is often possible to construct a lower-risk portfolio with attractive returns from very few assets.

The conventional wisdom is that low-risk allocations with good returns are not possible. However, this line of thinking arose from an era before near-zero-cost trading and unlimited computing power. These two factors alone create abundant opportunities for new investment strategies that would not have been fruitful just a few years ago, offering family wealth managers a wider range of appealing investment options.

Types of Algorithmic Strategies

Each type of algorithmic strategy represents a philosophy of market trading that often has existed for many years of traders manually entering trades at the computer. These strategies can be summarized by their adherent’s heuristics, such as “buy low, sell high” for value investors or “let your winners run” for momentum systems.

Family offices often find themselves flummoxed by the contradictions offered by the philosophies of different algorithms and wonder which is the ideal allocation strategy for them. However, the question itself reveals a general misunderstanding of markets because markets often shift through phases where one strategy may outperform another at different times. For example, a very strong market advancing parabolically presents few opportunities for a mean reversion system. In contrast, a sideways market may present no end to opportunity.

Momentum

Momentum strategies are rooted in the concept that strong markets will continue to be strong. Mathematically, many models assume that the returns for each time period follow a normal distribution and are independent of one another; in practice, this is rarely the case. Instead, markets tend to exhibit hot and cold streaks, not unlike a gambling table. It is understandable why this would happen when one considers that markets gaining momentum attract investors inclined towards higher risk. Thus, as excitement around an offering increases, investors pile on in hopes of making their fortune. A momentum strategy seeks to capitalize on this tendency.

Mean Reversion

A mean reversion strategy seeks to create small, consistent wins for the investor based on how markets suffer from minor mispricing over the short term. The mean reversion system scanning for these anomalies opens a position with the expectation that the asset’s price will revert to the mean. This type of system tends to work well when the market is in an accumulation phase and is range-bound.

Statistical Arbitrage

Statistical arbitrage systems calculate the relationship between the price of an asset and other data. For example, real estate has a strong inverse relationship with mortgage interest rates, so a statistical arbitrageur may design a model to measure the impact interest rates would have on real estate and trade the expected change in value based upon the change in interest rates.

Correlation in Algorithmic Allocation Systems

Investors responsible for family wealth allocation should be aware of a significant benefit when designing portfolios from an algorithmic allocation strategy—correlation can be designed into the system, even down to zero when diversifying amongst algorithms. This is desirable because there is often some level of correlation when diversifying amongst holdings to manage business risk or allocating amongst assets to manage market risk. For example, real estate allocations, stock allocations, and bond allocations are all sensitive to interest rates and changes in monetary policy, so they will be correlated to some degree, even if it is negative.

However, even a negative correlation is undesirable because there is never a time when all assets increase in value in parallel. Therefore, it is preferable to have uncorrelated instruments so that several can succeed in parallel without creating drawdowns in the others, thereby reducing portfolio friction.

Taxes and Fees in Algorithmic Allocation Models

A family office portfolio manager must take care to ensure that a potential increase in fees for an algorithmic allocation does not outweigh the benefits of the investment. These algorithmic investment strategies tend to be most useful for assets where volatility is high enough to overcome the hurdle posed by taxes and trading fees. While taxes rarely decrease, trading fees have converged to near zero due to the fees brokerages now receive for selling their order flow to liquidity providers. This poses an advantage for algorithmic trading systems that an illiquid market with high fees would not.

For example, an asset that offered a regular return of 5%, 1% trading fees, and no market price volatility would be a poor candidate for these techniques, whereas a cryptocurrency token with 90% monthly volatility may be a better candidate even after fees and taxes are considered. Regardless of algorithmic strategy, increases in volatility tend to present more opportunities to manage risk and, thus, more outsized return opportunities for the algorithmic investor.

Many family wealth managers choose to hold these types of investments in retirement accounts, private placement life insurance, or other deferred tax vehicles. This has the dual benefit of allowing the algorithmic strategy to compound without immediate taxation. These structures also tend to have long time horizons, further benefiting an algorithmic strategy.

Measuring Success with an Algorithmic Allocation

Measuring and tracking success for an algorithmic allocation poses certain challenges. Most investors’ intuition suggests comparing arithmetic returns. However, that can skew toward buy-and-hold techniques because drawdowns are not weighted heavily enough. A commonly utilized metric that helps compare a buy-and-hold strategy with an algorithmic one is CAGR.

Simulating Returns on Algorithmic Strategies

However, to understand the risk-reward dynamics of an algorithmic trading system, simulations are highly useful and provide exceptional insights.
When the portfolio manager employs a Monte Carlo simulation, they are running hundreds or thousands of simulations and observing the probabilities of each range of outcomes. This is important because markets are non-deterministic and probabilistic, as daily inputs are never the same. Even if they were, investors’ reactions to the fundamental data, news cycle, or price data would be different on any given day. Applying probabilistic analysis creates parity between these multiple factors: the speculative nature of the investment instrument, the statistical nature of the algorithmic system, and the reality that portfolio managers endure.

Risk-Adjusted Returns and Total Returns

Ultimately, the gold standard for an algorithmic system is its risk-adjusted return. Algorithmic trading system designers frequently track their Sharpe or Sortino ratios, which are quantitative methods used to compare the relative risk-adjusted return of an algorithmic investment model.
Rather than use the flawed arithmetic return, the Sharpe and Sortino ratios assume that market volatility is equivalent to risk. Both methods compute the returns and risk expected for each instrument and calculate the reward ratio expected for a given level of risk. This is useful for comparing trading systems, but the risk most investors feel is not the standard deviation of daily price changes—it is the peak-to-trough drawdown. In 2008, when many investors’ portfolios were down 50% or more, they felt the market would never recover and sold at the bottom. Investor behavior provides many examples of these irrational tendencies, supporting the thesis that their perception of risk is more related to total drawdowns than standard deviation.

Knowing that investor behavior is driven by total drawdowns, the best ratio may be something more akin to the Calmar ratio, where a ratio is calculated that describes the returns of the algorithm in terms of reward divided by risk. In this case, it is useful to compare algorithmic trading systems by dividing their expected return by their maximum drawdowns. This is a simple ratio to calculate and supports the investor’s intuition.

Finally, the algorithm must be compared in terms of total return. For example, if $10,000 were invested in the algorithmic strategy, how would the investment have fared compared to a benchmark? The simple answer is that if the algorithmic approach provides reduced volatility, reduced correlation, or improved returns, it may have a place in the portfolio.

Conclusion

An algorithmic allocation can complement many family office investment portfolios if capital is deployed artfully. When considering an algorithmic allocation, ensure that the investment fills a specific portfolio need. Too often, investors follow their deal flow and invest in highly correlated strategies that they understand too little. Family wealth managers should strive to curate their investments and understand how each investment’s strategy fits in terms of:

Exposure to business risk

Exposure to asset classes

Algorithmic strategy family

Correlation with other approaches

With that information, they can confidently allocate to algorithmic strategies in compliance with their investment plan.