Safety First: The Portfolio Theory That Got There Before Markowitz Was Famous

In 1952 two papers appeared that founded modern portfolio theory. One was Harry Markowitz's "Portfolio Selection." The other was A. D. Roy's "Safety First and the Holding of Assets." Markowitz won a Nobel Prize. Roy became a footnote, though Markowitz himself later generously wrote that Roy had equal claim to being the father of the field.

The two papers reach overlapping conclusions from entirely different starting points, and Roy's starting point is arguably the more honest one about what investors actually want.

The problem: nobody actually maximizes utility

Markowitz's framework assumes an investor who trades off expected return against variance according to a smooth preference. It is mathematically clean, and it is also a slightly odd description of a human being. Nobody wakes up in the morning thinking about their marginal rate of substitution between mean and variance.

Roy, who had spent the war doing operational research and had watched a great many people confront the possibility of ruin, thought people worry about something far more concrete. They worry about disaster. They have a level below which things become catastrophic: the pension does not cover retirement, the endowment cannot fund the university, the firm goes bankrupt. Everything above that line is negotiable. Everything below it is not.

So Roy proposed a different objective entirely. Rather than maximizing some abstract utility, the investor should minimize the probability that their return falls below a specified disaster level.

The key idea, via analogy

Imagine crossing a river on stepping stones. A utility-maximizer would think about the average dryness of their feet, weighted against the variance of dryness. That is not how anyone crosses a river. What you actually care about is the probability of falling in.

You would happily accept damp shoes on every stone if it meant you would never plunge into the water. And you would reject a route with beautiful dry stones if one of them wobbled dangerously. The objective is not "maximize expected dryness minus a penalty for its variance." The objective is: do not fall in.

Roy's investor thinks the same way. Pick a level of return, or of wealth, that would constitute a disaster. Then choose the portfolio that makes the chance of hitting that disaster as small as possible.

Now here is where it gets beautiful, and where it connects to everything else in finance. If returns follow a roughly bell-shaped distribution, minimizing the chance of falling below a disaster level turns out to be mathematically identical to maximizing a very familiar quantity:

Expected return, minus the disaster level, divided by the standard deviation of returns.

Look at that. If your disaster level is the risk-free rate, that expression is the Sharpe ratio. Roy derived it fourteen years before Sharpe, from a completely different motivation, and it appears in the literature as Roy's ratio or the safety-first criterion.

That is a striking piece of intellectual convergence. Two people asking different questions, "what is the best risk-adjusted return?" and "how do I minimize my chance of ruin?", arrive at the same formula. The reason is that the number of standard deviations between your expected outcome and your disaster line is exactly the thing that determines the probability of crossing that line. Maximizing the distance to disaster, measured in units of volatility, minimizes the probability of disaster. They are the same problem wearing different clothes.

Why it mattered

• It reframed risk as the chance of ruin, not as wobble. This is closer to how real institutions think. A pension fund does not care about volatility per se. It cares about failing to meet its obligations. Roy's framing is the ancestor of shortfall risk, of liability-driven investing, and of the entire "what is my probability of not making it" school of thought.
• It anticipated the Sharpe ratio. Roy's ratio is essentially the Sharpe ratio with a general threshold in place of the risk-free rate, and it arrived first.
• It is the intellectual root of value at risk. The idea of naming a loss threshold and managing the probability of breaching it is precisely the logic behind VaR and, in refined form, expected shortfall. Regulators now enforce that logic.
• It respects how people actually feel. Behavioral research consistently finds that people evaluate outcomes against reference points and care asymmetrically about losses. Roy built that in from the start, decades before prospect theory made it official.

The honest limitations

• It leans hard on the normal distribution. The elegant equivalence between "minimize disaster probability" and "maximize the ratio" holds when returns are roughly bell-shaped. Real returns have fat tails, and in the tails is precisely where a safety-first investor lives. This is a serious weakness, and it is the same weakness that afflicts value at risk.
• The disaster level is arbitrary. Who chooses it? Move the threshold and you get a different optimal portfolio. The theory does not tell you where to draw the line, and the answer depends entirely on it.
• It only cares about the probability of disaster, not its size. A one percent chance of losing half your money and a one percent chance of losing everything score identically. That is a real flaw, and it is why expected shortfall, which cares about how bad the bad outcomes are, eventually became the preferred regulatory measure.
• It gives up on the upside. By focusing entirely on avoiding a floor, the criterion is somewhat indifferent to how much you gain above it, which can lead to portfolios that are needlessly timid.
• Markowitz's framework generalized better. Whatever its philosophical shortcomings, mean-variance optimization plugged directly into the equilibrium theories that became the CAPM. Roy's approach, being about individual thresholds, does not aggregate as neatly, which is a large part of why it was the road not taken.

The one-line takeaway

Roy proposed that investors should minimize the probability of catastrophe rather than optimize a tradeoff, and showed that doing so means maximizing the distance between your expected return and your disaster level measured in standard deviations, which is the Sharpe ratio, discovered fourteen years early and for entirely different reasons.