The research viewpoint
For new and existing fund investors alike, PE is complex to navigate. With a wide dispersion of returns, a need to manage capital that will fund commitments over time, chunky, unpredictable distributions, and a decline in the persistence of returns from managers, the asset class can pose difficulties for LPs seeking to reach and maintain target allocations and predict the risk-return profiles of their investments.
These conundrums were what originally drove Yale University endowment investment officers Dean Takahashi and Seth Alexander in 2001 to create an approach that provides point estimates of cash flows and net asset values (NAVs) for a portfolio of PE funds. The model is still used by many LPs today, but its accuracy relies on an investor having enough sophistication to derive a set of assumptions to feed into the model.
Of course, the industry has moved on significantly since the early 2000s. Not only has it grown, but it has evolved to include structured financial products backed by PE, such as collateralised fund obligations and retail offerings, among others. The need to analyse new, more complex investment products, combined with an increase in historical data and computing power, led a group of researchers at Bella Private Markets, including Harvard Business School professor Josh Lerner, to revisit the Yale model.
“We wanted to keep the original model’s criteria – that it should be simple, capable of incorporating and responding to real data and of analysing the impact of various scenarios, and that it should be useful for a variety of asset types,” explains Bella’s chief operating officer, Alex Billias. “But we also wanted to solve some of its limitations. One issue is that even if LPs have data, they need to make assumptions, and the other is that you only get one outcome. The simplicity of the original model is therefore a double-edged sword because an inaccurate assumption could lead to a false outcome.”
The researchers developed a new simulation-based model that uses historical fund-level cash flow data. The model does not require users to derive assumptions as inputs; it requires only known cash flows and current valuations from the user’s portfolio of funds.
“Using this portfolio data as a starting point, our model then picks similar historical funds and traces what would have happened if the LP’s funds had evolved along the same trajectories as the historical funds,” explains Billias. “So, if you have a portfolio in 2023 with funds that are between three and eight years’ old, our model will ‘shift’ the portfolio back in time to, for example, 2011, and pick out three-to eight-year-old funds as of 2011 to match the vintage profile of the actual portfolio. It then traces the evolution of the historical funds’ cash flows and valuations. By repeating this process many times
over many historical periods, you can capture a range of outcomes that are mapped and scaled to the size and make-up of your own portfolio.”
Users can then consider a range of scenarios, such as how funds following a certain strategy might perform if they are hit by a downturn or shock. It can also help investors to see the effect of rebalancing their portfolios or making other adjustments. “It’s a probabilistic model as opposed to the deterministic approach taken by Takahashi and Alexander,” says Billias. “So rather
than providing single point estimates, we can provide a range of outcomes with probabilities of 10%, 20%, 30% and so on, based on the trajectories of similar funds in the past.”
The researchers say the model can be used in a variety of ways – to build LP portfolios, manage liquidity, develop funds of funds, in securitisation and structured finance and, potentially, in other private markets, such as infrastructure and real estate.
So what are its limitations? “One drawback is that, because it is using historical data, it can’t account for new developments we haven’t seen before,” says Billias. “So, it can’t tell us with historical certainty what effect using subscription lines of credit would have on capital calls, or how NAV lending might affect the portfolio’s return profile, because these are relatively new innovations. But, using the historical output as a baseline, you can layer on different assumptions and adjustments to account for these.” However, he adds, there are likely to be refinements to the model over time as more data becomes available and these market changes can be quantified.
“It could be a really useful tool for making something that is otherwise very difficult in PE and venture capital much more accessible,” concludes Billias. “After all, going back in time to see what happened in the past seems to resonate more with LPs than a multifactor regression model.”
Alex Billias