Don't use Vol to measure risk. Please.
Volatility, beta, value-at-risk (VaR), conditional VaR (cVaR) and correlation are highly interrelated risk measures; all based on linear regression. With no other theoretical foundation discovered since the origins of mean-variance analysis in 1952 and VaR in 1987, these calculations have been the only ones used for aggregating risk.
Despite many crises over the last 30 years showing that they fail to indicate ex-ante portfolio risk when crisis hits, they are still used actively by every institutional investor today.
Phrases similar to:
“all correlations went to one, what could we have done?”
“volatility just spiked suddenly”
“there was no price, all the bids disappeared”
“it was a one in a hundred-year event”
“we were driving looking in the rear-view mirror”
are commonly heard in a crisis, accompanied with metaphorical hands thrown into the air in resignation (and large synchronized losses).
Perversely, these backward-looking risk measures play a central role in encouraging the next crisis:
Prior to 2008, the seemingly low VaR and high ratings of asset-backed securities and other securitized credit allowed banks to accumulate leverage through credit derivatives, while keeping within Basel VaR-based capital requirements. This excessive leverage caused the collapse of Lehman Brothers and ultimately the Global Financial Crisis.
In our current cycle, the seemingly low volatility of private equity allows investors to increase allocation to illiquid positions in their portfolios seemingly without “risk” of short-term impairment. If we choose to use volatility as the main measure for top-level portfolio risk, then this dynamic becomes self-perpetuating.
Privates are missing entirely from the quantitative risk management equation
The lack of mark-to-market for privates means that no commonly used risk system aggregates specific risks from these illiquid positions. Public index proxies are typically attempted, but the whims of market sentiment and capital flows mean that the selected proxy price movements do not reflect the experience of holding these illiquid positions. They are then often ignored and ultimately dropped from the aggregation analysis.
Common risk systems today, including all those used by the largest institutional portfolios such as Aladdin Risk, Bloomberg PORT, Barra One and RiskMetrics, are unable to account for securities without a public security underlier or public proxy within its portfolio risk aggregation. These risk systems all use the same linear regressions to first calculate correlations and betas.
Regression is meaningless without strict periodic pricing data – monthly at a minimum – which immediately excludes all privates. Even for the majority of listed securities, market pricing provides highly misleading input for the risk system due to low, or at times zero, traded volume.
These correlations and betas in turn are the primary parameters used by all risk systems for calculating their reports, such as: (i) stress tests, (ii) Monte Carlo simulations, (iii) portfolio volatility, (iv) portfolio VaR and (v) portfolio cVaR.
If the correlations and betas are based on incomplete and inadequate data, as they are, then the output reports from these systems are equally incomplete and inadequate.
Expected Co-Drawdown (CoDD) has been developed to replace correlation
领英推荐
Expected Co-Drawdown (CoDD) is a measure formulated and formally published by the founder and CEO of FountainArc in the peer-reviewed Institutional Investor Journals, referenced in the following paper:
Lee C. “Expected Drawdown Management: An Ex-Ante, Long-Term Approach to Portfolio Construction” JWM (2016) https://doi.org/10.3905/jwm.2016.18.4.065
CoDD is central to the Expected Drawdown Management (EDM) theory described in the paper. The use of CoDD and Expected Drawdowns to manage portfolios is unique to FountainArc’s clients globally.
CoDD is simple in concept, but at the same time, it profoundly changes our ability to quantify and aggregate ex-ante risk like-for-like across every investment, whether liquid or illiquid.
As the name suggests, CoDD measures the ex-ante possibility of coincident peak-to-trough cumulative loss (i.e. Drawdown) between different investments. Unlike correlation, which is defined as a backward-looking statistical measure, CoDD is crucially defined as a forward-looking Bayesian measure.
CoDD can be interpreted as the probability of suffering cumulative losses from different investments in the future at the same time.
A CoDD of 1 means two investments will always have synchronized cumulative losses and a CoDD of 0 means two investments will never have synchronized cumulative losses. Most actual investments will have CoDDs somewhere between 0 and 1. Please refer to the above referenced paper for all the mathematical details and in-depth method testing.
Importantly, CoDD is based on Expected Drawdown, and drawdown is a cumulative measure, unlike volatility, which is a strict periodic measure. For this reason, CoDD and Expected Drawdown do not require strict periodic pricing and are highly tolerant of misaligned and missing returns.
They apply equally well to quarterly, semi-annual or even annual valuation data, and can handle irregular valuation points typically found in private equity. These data would be entirely impossible to use in calculating volatilities and correlations.
Further, CoDD and Expected Drawdown do not even require pricing data at all. Their nature of being forward-looking (Bayesian) by definition, and also being fundamentally intuitive, it is?possible to estimate and concretely justify CoDD and Expected Drawdown parameters using logic and research analysis alone. This would again be impossible with volatilities and correlations.
FountainArc provides the full quantitative system for fast implementation of EDM
FountainArc Technologies provides a unique enterprise cloud solution for large institutions to deploy automated EDM theory and CoDD analysis across their portfolios and mandates. The solution is intuitive to use, and not only calculates and manages the portfolio-level Expected Drawdowns, but also automates across the processes of collecting data, maintaining data and modelling parameters.
Automation is such that onboarding is fast. Working initial models and institution-wide system are deployed potentially within one week, and fully refined parameters and accurate modelling are potentially deployed within one month (based on the typical number of material holdings for an institutional portfolio).
The system is entirely collaborative, facilitating the contribution of research and data from multiple team members using real-time collaboration capabilities throughout.
No software installation is required as the system is managed and hosted by our lightning-fast servers from a server building fully owned by FountainArc. The system is securely accessed through 256-bit encryption using any of Windows PCs, Mac laptops or Android/iPad tablets on the cloud. With the entire institution collaborating at any time and from any place, the system will apply the institution’s best collective intelligence and management to their entire portfolio.
Contact FountainArc to begin fully managing your portfolio
Coordinate with Rachel on +44 207 770 6899
or email [email protected]
Head of Investments & Financial Life Solutions
5 年Interesting. Would be curious to see more of the forward looking methodology. Sounds like a combo of ex ante correlation/VaR, which essentially is often a conditioned historical path through ewma or else. Or it’s the result of some form of economic scenario model which then is as good and useful as your economic prediction accuracy and model calibration. Not a revolution from asset vol/cvar but a nice different lens. Side note, anyone using vol on private markets though (as a colleague once said) should be “shot”. And you get around this very easily using factors. You should also be able to address the “Co-“ by turning all your assets into factor exposures, because there, you have implicitly extracted all the possible relevant inter-dependency between assets. You can still have CoDD between factors but it’s a much simpler problem. Finally, I could not agree more that plain historical vol is only an “after-the fact” measure - so useless in predicting future. unless you manage to identify levels where the probability for it to persist in that stage is high (the mathematicians would tell it’s called an ergotic matrix you need). Wish you the best Chi in your endeavour!
B2B Podcast Producer @ orama.tv | ??Host Investology: re-think investment management |?? InvestOrama: 20k+ on YT | Into Swimming, Surfing, Parenting
5 年This is brilliant. I always felt that volatility was an inadequate proxy for risk, and although many people would agree, nobody came up with a solid alternative. I will definitely look further into CoDD
Management Consultant ESG Sustainability & Index Design
5 年This is the most intelligent alternative model to traditional risk metrics I have seen in my over 25 years in the industry. Since my early days, I always professed that volatility is NOT a measure of risk. As an investor, the only risk that matters is the possibility or probability of losing principal - full stop. There are only two prices that matter - what you paid and what you sold at, especially if considering pension investment. Long term return matters much more than month-to-month returns. And yet the industry obsesses about daily, weekly, monthly and quarterly performance. This is particularly painful for real value investors waiting for mean reversion which isn’t a weekly occurrence! My first experience as a fund manager was painful as my clients were focused on 3-5-10 year return while the consultants and sponsors were obsessed with quarterly performance. I came to the conclusion early on that it is impossible to consistently add alpha over the long term if you obsess about short term returns. In any case, while this is a solid model for irregularly-priced assets, I think asset managers AND clients should refocus and redefine their tolerance of risk in terms of loss of capital for all assets. Well done !
Pro Jock at Scottsdale National Golf Club, L.L.C.
5 年If you want a hedge build a garden!