Manish Khatta joins forces with HiddenLevers co-founder Praveen Ghanta to launch GuardRails, a risk analysis play that rejects raw 'risk numbers' -- a 'game-changer' in the scale-minded RIA space, analyst says

Maybe the future is advisors building real tech? Specialized tools, unique to a firm's needs will be worth more to larger RIAs than one-size-fits-all solutions.

People generally don't understand "risk models" in the least. For example, the question that should be addressed in this article is "what is the dataset that you used to construct your risk model?" An answer of "we used data over the past 20 years" is really setting you up for disaster. The last 20 years have been an unprecedented time of central bank loose policy - money printing. And, of course, risk (as defined by asset price variance and/or ultimate drawdown) has been dampened by this unsustainable central bank policy. So, how well is a "maximum drawdown" model expected to work in an environment of fiscal spending conservation? And if there is no fiscal spending conservation, then isn't the biggest risk that ultimately there are no buyers for bonds spewed forth by the serial money printers sometime in the next decade? And that's a risk that isn't captured at all. Not by variance based models, and not by maximum drawdown models. It's called "risk" for a reason.

Risk models are 100% math. Most people fail to understand even the most basic math, and therefore have no understanding of risk models. When was the last time an advisor pulled up Excel and calculated a covariance? Most can't even explain the concept. Max draw down is one way to measure risk, is it the best? That is up for debate. Opining about central bank policy sounds more like a CNBC talking point. It it represents one type of risk, but it does nothing to help you MEASURE risk. At least Max Draw down is something you can use to compare a cross section or securities / portfolios.

"Risk models are 100% math". I agree with this statement. But like any math, the answer you get is dependent on the inputs you provide. If one is using a biased sample (ie. last 20 years of historical data) and the forward looking data is outside the sample, the VALUE of your risk estimate is essentially zero. I'd agree that "maximum drawdown" IS a better definition of risk than "volatility", but both will prove of limited value in managing portfolios in "outside sample" environments...and isn't that the whole point of creating a risk model - to recognize and mitigate risk? I'd argue that there are insights that risk modelers could use to provide better estimates of expected risk than wholesale sets of historical data - these include what I'd call "regime shift" identifiers and bayesian adjustments. A simple overlay would be to recognize regime shifts and then re-engineer the covariance matrix to account for the fact that in down markets, correlations of risk factors invariably move closer towards 1 and variance terms increase by say 30%. Naive perhaps, but worth exploring.