Unlike traditional investing forecasting risk and return becomes challenging when dealing with alternative investments. These have shorter data histories, expose investors to new kinds of risk and, in some cases, offer limited liquidity. A method is to attempt to purposely remove the differences between alternative and traditional asset classes so that investors can evaluate them in a like-for-like fashion. The approach is threefold. First, to modify the historical characteristics of each alternative asset class to ensure that they are comparable to fixed income and equity. Second, to develop an integrated set of return forecasts using historical relationships among alternatives and traditional asset classes. Third, to add constraints to capture preferences of typical investors and account for issues such as shorter data histories for alternative classes.
Alternative asset classes differ from traditional asset classes in several ways, making it difficult to properly characterize risk. The differences, as well as our approach, built off our previous work at CAI, to correcting them, include:
• Risk in non-traded, illiquid assets. One of the most difficult problems investors face is how to compare the risk in traded assets, such as fixed income and public equities, to the risk in non-traded assets, such as real estate and private equity. The fundamental challenge is that estimating risk from non-traded assets requires an investor first to “mark to market” the valuations. ActiveAllocator has refined dated methodology that incorporates information of various kinds in order to estimate risk on a basis comparable to that of liquid assets. This methodology indicates that both risk-adjusted returns and diversification benefits from these asset classes are less than previously believed.
• Restrictions on tradability. Many alternative assets set restrictions on when the investors can exit the investment—what is called “tradability.” These restrictions affect the ability to re-balance, respond to new investment opportunities and meet unforeseen cash requirements. In exchange for these limitations, investors are paid a premium for locking up their assets in addition to the premium they can expect in the long run for taking risks. The failure to appropriately account for this premium, however, can lead investors to inappropriate portfolios. Therefore, a portfolio construction approach that accounts for these restrictions by incorporating tradability into the traditional risk/return framework is usually better.
• Reporting biases. Because industry returns are usually reported for a sample of funds, the performance of many funds—which often, but not exclusively, turn out to be those funds that perform poorly—do not appear in published indices. This reporting bias has the potential to artificially inflate reported performance. ActiveAllocator’s approach allows investment advisers to correct for these biases to returns.
• Serial correlation in hedge fund returns. As is the case with the highly illiquid assets such as real estate and private equity, certain hedge fund strategies such as distressed debt are characterized by pricing distortions in reported returns. This can lead to an understatement of risk. Our methodology corrects for serial correlation which refers to correlation of returns over time, and typically occurs when assets are traded infrequently. We, as others have found, too discover that removing the effect of serial correlation increases historical risk compared with reported values.
• Strategy drift. The fact that alternative investments are often unregulated, active strategies means that fund managers may change their investment focus over time to seek higher returns, introducing potential risks that investors may not have known that they were taking. This risk can be mitigated by aggregating hedge funds into more stable units, thus attempting to limit strategy drift at the portfolio construction stage.
• Tail risk. Unlike traditional investments, some alternative investment strategies are peculiarly subject to the risk that returns will be significantly below their average. For most investors who are particularly concerned about limiting losses, this potential for significantly below-average returns means that traditional measures of risk, such as volatility, might be inappropriate. We build on a framework that allows investors to construct portfolios of alternatives that explicitly account for downside, or tail risk.
• Fees. Reported index returns are net of manager fees. However, many investors access alternative investment products, such as hedge funds or private equity, through a fund of funds. This introduces a second layer of fees that the reported returns do not capture. As a result, reported returns actually overstate the value captured by most investors. We account for this by stripping out fund of fund fees from the historical data.
• Appropriate time frame. Over short time periods, returns for an asset class can depend heavily on that asset class’s stage in the investment cycle. For example, equity returns look very different in bull markets than in bear markets. In order to properly characterize the risks and returns of an asset class, investors must analyze these characteristics over a full investment cycle. However, cycles for certain alternative investment classes, such as real estate or managed futures, are longer than those for equity and fixed income. We adjust the length of historical data used for each alternative asset class depending on the length of the investment cycle.
After modifying the historical data to account for the issues listed above, ActiveAllocator uses the modified data to compute the risk characteristics – standard deviations and correlations – of each asset class. The differences can be fairly dramatic.
Using historical return forecasts to construct portfolios is problematic, because return characteristics are much less stable than risk characteristics. This problem can be circumvented, however, by recognizing that relative returns among asset classes are far more stable. For example, while equity returns may vary substantially over different time periods, the spread between equity and cash is far more consistent over time.
Our approach starts with the forecasts for traditional asset classes described in the previous section. From these basic building blocks, listed above, ActiveAllocator constructs forecasts of average returns for alternatives. For each of the alternative asset classes, we begin with a spread-based method, mirroring the approach taken on the traditional side, which calculates the differences between returns in alternative asset classes and related traditional assets.
For example, to calculate a forward-looking return for each hedge fund strategy, we begin with the spread to cash as a basis. These spreads, as it turns out, have been more stable over the period for which hedge fund return data have been collected. We then estimate the relationship between this spread and other factors that drive hedge fund returns, in particular exposure to equities and exposure to credit. The results of this analysis allow us to tie nominal hedge fund return estimates to our forecasts of the other asset classes by incorporating the forward looking forecast for cash, equities and credit. Following our earlier discussion, we then adjust these forecast spreads for a number of other factors—such as survivorship and selection bias, operating expenses which might be incurred at a fund of funds level and indicative fund of funds fees.
For managed futures, we also use the spread to cash as a basis for forecasting. Here, again, we must adjust for the fees incurred by various investors. Managed futures return however are often already net of fees. As separate accounts, they are generally reported net of the fees a fund of funds manager would take, for example. As a result, it might seem as if no additional adjustment for fees is necessary. Notwithstanding that point, however, managed futures fees will still differ across various investor types. We therefore make an additional adjustment to the spread-based forecast to account for fees that might be incurred by investors who invest in smaller lot sizes.
For the private equity classes, including leveraged buyouts and venture capital, we calculate the spread between returns in each of these categories to the public equity market. We then add this spread to the forecast public equity return to arrive at the forecast for venture capital and leveraged buyouts. Finally, for real estate, we apply a similar methodology that not only accounts for interest rate movements, but also for vacancy rates and net operating income. This assembly of return forecasts creates a complete set of parameters for use in constructing portfolios: risk, return and correlation.
As in any asset allocation exercise, it is necessary to consider constraints on particular asset categories. This is because models are only approximations of reality, and do not capture the full set of criteria on which investors base decisions. Some of these criteria include:
Tradability – As described earlier, many alternative assets set restrictions on when the investors can exit the investment, which imposes a cost on the portfolio. Therefore, we allow investors to constrain the amount of illiquidity in their portfolios based on their cash flow requirements.
Tail risk – Although, in our earlier discussion we have attempted to remove the unaccounted for tail risk, some residual tail risk might still exist. All things being equal, investors should constrain allocations to asset classes with significant tail risk, as tail risk increases the probability of significant losses. Therefore, our optimization algorithms take negative skewness and excess positive kurtosis effects into account.
Confidence – Even after modeling each asset class, investors will have more confidence in the forecasts of some asset classes relative to others. This can be due to the length of data with which our research methodology operates. Whereas, at most, alternative asset classes tend to have approximately twenty years of data, data at least four times that length are available for many publicly traded assets. Furthermore, some asset classes experience more rapid change in their structure than others. For example, the hedge fund industry has been characterized in the last few years by a decline in returns and more recently an outflow of capital. In contrast, equities and fixed income have been much more stable. Therefore, our optimization algorithms apply levels of confidence to different forecasts.
Market Capitalization – Another indicator that is generally used as a basis for ensuring diversification is the market capitalization of the various asset classes. This is because an asset class’s relative market capitalization reflects the market’s consensus regarding an appropriate allocation. If the market capitalization of a particular asset class is small relative to other asset classes, investors should constrain allocations to this class, unless their beliefs regarding future performance differ significantly from the market consensus. Therefore, our default constraints take this into account, but we allow investors to override this.
Risk Tolerance – In general, investors who are more willing to take risks will be more willing to bear unaccounted for downside risk and uncertainty about forecasts. These investors may generally place less restrictive constraints on their portfolio. Therefore, our defaults allow investors to specify upto three bands of risk tolerance.
A pre-requisite for constructing any great portfolio is identifying skilled managers that an investor believes can consistently generate alpha. Although investors intuitively recognize the value of alpha, they have historically lacked the tools required to build portfolios that include active managers. ActiveAllocator is trying to address newer challenges by designing a rigorous, integrated and, perhaps most importantly, practical framework for active investing. Our extensive testing provides evidence that this framework can significantly improve upon more traditional portfolio construction methods. The approach is unique in that it:
- More accurately measures manager alpha and beta on historical basis
- Explicitly forecasts manager performance by combining historical data with other information
- Quantifies unique risks of active managers
- Accounts for these unique risks when constructing portfolios
While the details of this approach are outside the scope of this note we have tried to summarize the general principles here. We hope they benefit investors even if they use other qualitative portfolio construction techniques.