How to Select a Portfolio Optimizer that Supports Your Investment Priorities

Whether you are a systematic asset manager, an asset owner, or a wealth manager, a portfolio optimizer is a powerful tool to research investment ideas and put them into production. Given the many optimizers in the marketplace, how can you distinguish them and discern the signals from the noise in your evaluation process? In this article, we will examine the key factors to do that.

Why Use an Optimizer in The First Place?

Optimizers enable you to model your investment strategy, account for mandates/constraints, and incorporate transaction cost estimates and tax implications.

By systematizing and replicating your portfolio construction workflow, you gain consistency in your investment process. This can help prevent emotional decisions as well as reduce the chance for human error from manual processes. Incorporating a multi-factor risk model¹ can also enable you to uncover and neutralize hidden/unwanted factors that affect your portfolio.

Let’s walk through key points to consider.

Data Integration

A typical optimization will pull from several databases, fetching inputs and attributes such as:

• The risk model

• Symbols

• Prices

• Portfolio share quantities

• Fundamental characteristic values

• Sector grouping classifications

Gathering all this data can be a nontrivial process, as it may come in multiple formats from multiple sources and may need to be cleaned and formatted prior to use. An optimizer provider that offers a seamless data-integration process can reduce time spent wrangling data and free up more time for you to provide clients with strategic value.

Many clients also use proprietary alpha signals to guide their portfolio construction process. Being able to consolidate your optimization and alpha generation process onto the same platform can be more cost efficient and logistically easier to load your alphas into your optimizer application. This can lead to a more streamlined workflow and mitigate risks that can arise when transferring/translating data across platforms.

Platform Flexibility

There are multiple providers of portfolio optimization applications and risk models in the marketplace, each with their own sets of features and methodologies. Evaluating the offerings from multiple providers can be time-consuming; a single platform that enables you to choose the optimizer and risk model that best suits your needs can help streamline this process.

In addition, it is important to consider the additional workflow components offered on the same platform. If you are interested in evaluating potential factor signal performance over time, a portfolio backtesting application that incorporates an optimization engine is key. If you want to make trades to a live portfolio, the ability to export your trades in your OMS format can save you time.

To choose the provider with the best workflow solutions to meet your needs, you will want to evaluate the overall flexibility of the optimizer platform or provider. Here are key points worth considering:

Optimizer engine: Optimizers are available from many providers with a variety of features, functionality, and coverage of asset classes. A single platform that offers optimizers from multiple providers can provide one-stop shopping for evaluating which engine offers the best combination of tools for your workflow.

Multi-factor risk models: Also available from a variety of different vendors, they utilize different methodologies and cover different countries, currencies, and asset types. You will want to select the model (or models) that provides the broadest coverage of your investment universe and captures exposure to a broad set of explanatory risk factors that align with how you think about risk.

If your primary risk measure is volatility, consider either a fundamental factor (parametric) model or statistical model; if your primary risk measure is expected tail loss (ETL), you should utilize a Monte Carlo-based risk model.

A platform that provides off-the-shelf interoperability of risk models from different vendors with your optimizer engine of choice will allow you to easily test various models to determine the one that provides the best coverage for your portfolio.

Interface: Many users prefer a graphical user interface (GUI) with point-and-click applications to create optimizer documents. However, depending on the workflow, some individuals prefer using an optimizer programmatically, such as those performing backtests or executing large numbers of runs.

In this case, consider using an optimizer available either via API or through a JupyterLab-based application such as the FactSet Programmatic Environment (FPE). FPE enables you to easily combine code, rich text, charts, and mathematics into a single document.

Ecosystem: A provider that integrates its optimizer into a larger workstation can be compelling as it allows for greater connectivity and communication across apps. For example, while refining your optimization trade list, the ability to easily launch news or company reports for a particular stock will make your workflow more dynamic and efficient.

Scheduling: While you may want to perform some iterative tasks interactively, it’s more efficient to schedule longer simulations or regular production jobs to run automatically.

OMS output: If you want your optimizer to generate trades for execution, you need the ability to export those trades directly to your OMS. That efficient automation reduces the potential for human error by manipulating values in a spreadsheet.

Trade adjustments: Many users want to edit trades that are returned from the optimization engine, prior to sending them for execution. Perhaps the optimizer recommends purchasing a particular company’s stock about which a significant negative news story has just broken that morning. That’s information the risk model is not aware of, requiring user intervention to eliminate the trade. The ability to manually adjust trades within the application (rather than in Excel) is preferable as you can more easily review the impact this change has on the cash position, active risk, and factor exposures of your optimal portfolio.

Frontier analysis: Selecting the constraint bounds that provide the best results and remain within your investment mandates requires an iterative process of testing input values. An optimizer with the ability to natively construct a frontier of different optimal portfolios, by varying a particular input (e.g., the max bound of a turnover constraint), can make your research process more efficient and less time consuming.

Simulation: This entails backtesting your strategy over time with actual historical prices/data to see how it would have performed. That step can be quite valuable. It helps to identify the strategies/signals that have the best probability of capturing alpha. Conversely, it can help weed out those with the lowest probability of added value. A simulation app that also accounts for intra-period actions, such as dividends and spinoffs, for example, can make your simulations even more realistic.

Reporting: Access to robust pre- versus post-trade reporting is important because it confirms the impact of recommended trades on the portfolio. This includes:

• Asset-level and factor-level reporting

• Portfolio level values such as change in active risk, number of holdings, or information ratio

• Group values, such as change in sector or country weights

Reporting that can easily delineate the impact trades will have on your objective and constraint values are helpful in evaluating the optimization output.

For example, the following FactSet report shows the minimum/maximum and initial/final values for several constraints being applied in an equity optimization. It is constraining on several factor exposures, turnover, sector weights relative to the benchmark, and active risk.

Below is an example of an initial versus final summary report in Portfolio Analysis, the optimization performed with the FactSet Portfolio Optimizer. It shows several fundamental and risk-related characteristics for a balanced portfolio, with the risk and factor exposures measured with the FactSet Linear MAC Monthly Model 2 – MH.

Optimization Features

When evaluating optimizer features, consider two main categories: the strategy settings and the engine or solver that determines the recommended trades. Let’s take a deeper look at the key strategy settings, broken down into the following common groups.

Objectives: The objective defines the overall goal of the optimization, formulated in terms of minimizing and/or maximizing certain criteria. The engine uses your objective to determine trades that will add the most value to your portfolio.

The classic case is to employ a mean-variance objective, where the goal is to minimize risk and maximize expected return/alpha. Other common objectives include minimizing the level of transaction cost, maximizing the Sharpe ratio, or tilting toward a particular attribute. For tax-aware optimizations, minimizing tax liability is also a commonly used objective.

Constraints: Constraints impose hard bounds for your optimization. For example, you could define a maximum of 20% turnover, a minimum of 30 names in the portfolio, and a 4% cap on active risk.

Constraints can be grouped into several categories, such as linear (e.g., weight-based constraints), cardinality (e.g., trade threshold or number of names), or quadratic (e.g., a constraint on variance of a portfolio). Combinatorial constraints such as number of names are harder to solve than others. They require longer runtimes and potentially lead to less optimal outcomes. While your mandates may require such constraints, it is generally recommended to avoid including undue constraints in your strategy, as constraints often force the optimization engine to return a solution with a poorer objective value than could be achieved in the absence of any constraints.

Transaction cost: A viable investment strategy must hold up under real-world conditions, and real-world trading incurs transaction costs. Select an optimizer that supports transaction-cost modeling; at minimum, this should include general linear transaction cost. To create more nuanced transaction cost estimates, a piecewise-linear or a market-impact-based cost model may be worth considering.

While some settings such as a turnover constraint are relatively standard across all the optimizers in the market today, others are less common—and some are unique to individual providers. It’s important to review your investment policy statement or guidelines, as you will want to pick an optimizer that provides the features sufficient to capture and model your investment approach.

For example, if you have a mandate to target the beta of your portfolio to be within 1% of your benchmark, ensure your optimizer can target a weighted average/sum of a particular attribute. Or, if you use cardinality constraints in combination with an objective of maximizing the information ratio, filtering the available optimizers based on the criteria would be important.

Each optimizer also uses its own solver/algorithms to produce trade recommendations. As mentioned above, some settings make an optimization harder for the engine to solve than others—most notably, combinatorial, and non-convex constraints. When testing an optimizer, a good approach is to incrementally add constraints to your strategy and re-run after adding each of them to assess the impact on your objective values. For example, if your objective is to minimize active risk, you can view the change in active risk in your optimal portfolio before and after adding an additional constraint).

You can also set up the same strategy in multiple optimizers and evaluate the change in the objective values.

Cost

Optimizers are often priced on a per-user basis, but some workflows might be priced on volume or the number of portfolios. Certain add-on features may also require an additional cost, such as tax-aware optimization or a scheduling application. Some key financial questions to answer include:

• How many users at your firm need optimizer access?

• Do they require UI-based or programmatic access?

• Which features and workflows are need-to-have, which are nice-to-have, and which are not needed? For example, you may absolutely need the ability to backtest while targeting sector weights to be within a certain percentage of the benchmark; you might like the ability to target certain factor exposures against a secondary benchmark; or you might not need the ability to run a long/short optimization.

Answering those three questions will help you discern an optimizer within your budget that offers the most features from your wish list.

Risk Measure

The most common risk measure in portfolio optimization is volatility, typically measured with standard deviation versus a benchmark (although some may use absolute level of volatility). This is exhibited in the classical mean-variance problem, where one seeks to maximize return and minimize volatility to achieve the portfolio with the greatest amount of return per unit of risk. Assuming a standard-normal distribution of risk and return, this type of optimization is well-suited for linear assets—such as equities and fixed income securities—and is captured with parametric or statistical risk models.

However, in cases where non-linear assets such as options are present, or asset returns are non-normally distributed (with increased skewness and fat-tails), typical mean-variance optimization falls short. A better risk measure is expected tail loss, which seeks to capture the potential loss on the portfolio, conditional on the loss being greater than the VaR (Value at Risk). ETL is a risk measure supported by Monte-Carlo simulation-based risk models. An optimizer that supports ETL as a risk measure enables you to incorporate this type of analysis into your portfolio construction process.

Conclusion

Adding an optimizer for portfolio construction is an important decision that can lead to a more efficient and consistent investment process. To select the optimization provider best suited to your priorities, consider the one with the most flexibility and the features that best model your investment and risk-evaluation approach.

¹ Multi-factor risk models use multiple factors, which are characteristics that are correlated with asset returns, to attempt to explain and decompose portfolio risk, by capturing the volatility and covariance of individual assets in the portfolio to these factors. These models also capture stock-specific risk, which is based on the volatility of the residual asset returns. Such models seek to offer greater explanatory power than a single factor model would offer, such as the Capital Asset Pricing Model (CAPM), which seeks to capture a portfolio’s correlation with a single market factor (beta).

This blog post is for informational purposes only. The information contained in this blog post is not legal, tax, or investment advice. FactSet does not endorse or recommend any investments and assumes no liability for any consequence relating directly or indirectly to any action or inaction taken based on the information contained in this article.