Vertex Macro | Proprietary Trading: Truth and Fiction

Proprietary Trading: Truth and Fiction

Peter Muller

Peter Muller spent eight years building what he calls a “reasonably successful proprietary trading team.” In this article he discusses some of the core questions behind the strategy—without giving away the house edge.

Our team’s trading method is model-driven. Before we implement a trading strategy, we spend a great deal of time on research and backtesting. If you read an academic article on an asset-pricing anomaly, we have probably read it too, usually verified the study, and occasionally applied it—in modified form—to one of our strategies. For competitive reasons I will not say much about what we trade, how much capital we manage, or our track record. After all, I want the team to keep producing excess trading profit for a considerable time.

I am reasonably confident that our results are not entirely luck (though I sometimes wonder whether we are any different from the lucky monkey that happened to type Hamlet on a typewriter). As far as I know, our results are independent of market direction (we are market-neutral), independent of a liquidity premium (our positions are liquid), and independent of a skewness premium or other optionality (our upside and downside risks are symmetric). I also do not think our results come from lower trading and financing costs that an investment bank might enjoy internally (when we compute performance we assume we pay commissions and financing spreads). Of course, although we sit inside an investment bank, we have no access to information about the bank’s clients. Still, we may always be exposed to some risk factor we do not yet know.

Risk management

We manage risk in two ways. First, for each strategy we set limits on some or all of the following: capital usage, expected portfolio standard deviation, value at risk (VaR), position liquidity, and exposure to various predefined risk factors (for example, a rho limit constrains the first derivative of portfolio return with respect to a parallel move in interest rates). Incidentally, our client—the bank—also imposes aggregate limits on many of these variables, plus some of its own.

But the most important risk is that our models may stop working. To reduce that risk, we set a loss target for each strategy. If losses exceed the preset target, the strategy is reassessed and temporarily shut down (possibly permanently). This is not very different from how traditional proprietary traders are managed: “Go trade, don’t take too much risk, and if you lose more than x dollars we will shut you down.”

Our strategies are evaluated with return/risk metrics. For a symmetric, market-neutral strategy without significant tail events, the Sharpe ratio (SR) may be the best ex ante measure. SR is defined as the portfolio’s annualized excess return divided by the annualized standard deviation of that return. Our benchmark is cash, so measuring excess return is appropriate for our portfolios. For a long-only manager, the information ratio—excess return versus a benchmark—is more suitable.

When we assess past performance we also look at peak-to-trough drawdown (the largest drop between consecutive maxima and minima over the life of the strategy) as an additional risk variable. This helps find serial correlation in portfolio returns that the Sharpe ratio misses. It is also worth watching the share of expected gross profit consumed by expected transaction costs. The higher that number, the more we expect to lose once the model fails.

Our edge comes, at least in part, from opportunities created by institutional managers who overtrade. Their trades are often based on overestimating their ability to forecast investment returns, or on misunderstanding how transaction costs rise with size. Although these managers supply us with opportunities from overtrading, their strategies can still be entirely rational—simply because portfolio management contains large agency problems.

Incentives and performance

Investment strategies have a fixed capacity. When I add money to a strategy, my expected transaction costs rise, while the estimate of expected return before costs stays the same. Figure 1 shows this: once my marginal expected return (after transaction costs) falls below zero, adding investment only loses money.

Unfortunately, for most investors who delegate fiduciary responsibility to an investment manager, investor and manager incentives are not aligned. Almost all investment managers are paid as a percentage of assets under management. In mutual funds and most institutional portfolios, that means the primary incentive is to manage more money. Performance does help determine pay, but only because it helps retain assets or attract more assets.

I am not saying investment managers do not try to deliver superior returns. I am saying the economic incentive pushes many successful managers to the right of Figure 1. The problem is worse because of large errors in estimating the shape of the chart: most managers badly overestimate their ability to forecast asset returns. When someone wants to give you money to manage, it is hard to refuse them, and easy to believe you sit further left on Figure 1 than you actually do. (Note: as an asset manager grows, expected return may initially rise with assets, because extra compensation can be used to improve the investment process.)

Hedge-fund managers are paid both a percentage of the profits they earn and a percentage of assets under management. They are therefore less exposed to the agency problem above, but not immune. Asset-management firms usually charge only a management fee, and trade at 8–12 times earnings; hedge-fund managers also charge a management fee and, if they sell the firm, can still get 2–3 times earnings.

By contrast, proprietary traders usually earn only a percentage of their trading profits (that is our case). Holding more assets than we need does us no good, because we pay financing costs on every position. If the bank has sound risk-management controls, incentives line up well. Excellent risk management is essential to avoid giving traders too much free option value (think Barings). Another way to reduce the option value that the pay structure gives a proprietary trader or hedge-fund manager is to withhold part of compensation as a reserve against possible future drawdowns. That further aligns incentives and removes some asset-substitution problems.

A simple study suggests itself: rank investment-manager categories by the risk-adjusted returns of all managers in the category. It would be surprising if basic economic incentives failed to determine relative investment performance across categories. Proprietary traders have the purest incentives and perform best; hedge-fund managers come second; institutional portfolio managers come last. I believe this remains true even after including the well-known disasters in hedge-fund management and proprietary trading over the years. Accurate proprietary-trading performance data may be too hard to obtain for such a study to be feasible, but the hedge-fund performance literature that tries to estimate group-wide results is consistent with these claims. (See “Further reading” below.)

More simply: if your investment firm has a marketing department, you may not be a particularly good investor. See Figure 2.

Figure 2: Investment-firm politics

How to build a high-Sharpe-ratio strategy

How do you build a high-Sharpe-ratio strategy? Well, you will not get much advice from me—sorry. I will discuss two questions. First, should you try to build one truly great strategy, or combine a pile of decent ones? Second, how do short-term and long-term strategies differ?

In Grinold and Kahn’s book on active portfolio management (see “Further reading”), the authors describe the “fundamental law of active management”: a strategy’s Sharpe ratio is proportional to the number of independent bets the strategy takes times the correlation of those bets with their outcomes (Figure 3). To raise SR, you need more bets or higher forecast strength.

Figure 3: The fundamental law of active management (Grinold and Kahn 1999)

SR² ≈ N × IC² / (1 - IC²)

SR: Sharpe ratio. N: number of independent bets. IC: information coefficient (correlation of bets with outcomes).

In my view, refining a single strategy by increasing the number of bets inside it and raising forecast strength is far better than trying to stitch together many weaker strategies. For an investment strategy, depth matters more than breadth.

As a strategy develops, betting opportunities increase and the payoff per bet also increases. But a large transaction-cost barrier must be overcome before the strategy becomes profitable. Once that barrier is cleared, further improvements lever profit more effectively than the initial research. Those improvements are harder to get, but the effort is worth it. I know a proprietary trader to whom a shrewd investor offered extra pay if he focused on improving his original strategy rather than developing new ones in different fields. He eventually found ways to multiply the original strategy’s return several times.

If you add transaction costs to the equation in Figure 3, the relationship between forecast strength and Sharpe ratio is no longer linear. It looks more like Figure 4. (Interestingly, the experience of many proprietary traders and hedge-fund managers I know looks exactly like that chart, except the x-axis is “effort” and the y-axis is “return.”)

I would rather own a single strategy with an expected Sharpe ratio of 2 than a combined strategy with an expected Sharpe ratio of 2.5 built by stitching together five supposedly uncorrelated strategies each with an expected Sharpe ratio of 1. In the latter case you face the risk that the strategies are more correlated than you realize (think Long-Term Capital Management). Confirming that each individual strategy really has a Sharpe ratio of 1 also takes extra work.

Of course, where you dig matters. There are already established model-driven teams in convertible arbitrage, mortgage-backed securities arbitrage, futures trading, long-short equity statistical arbitrage, and options arbitrage. Some teams use more than one strategy, but to earn substantial, reliable profit in these fields you need to put in enough work to become one of the top players in each.

Short-term and long-term strategies

For many proprietary traders, an important choice is whether to focus on short-term or long-term strategies. Typically, the edge in short-term strategies comes from supplying temporary liquidity to the market, or from forecasting short-term trends produced by inefficient trading. Long-term models look at inefficiencies in asset pricing. How do these strategies differ in implementation?

Short-term investment strategies are popular because they often produce higher Sharpe ratios. If your average holding period is a day or a month, you have many more bets than if you hold for three months to a year or longer. On the other hand, short-term strategies often have capacity problems (it is easy to make a little money with them, hard to make a lot). They also require large investment in trading infrastructure, because fast, low-cost execution matters much more than in long-term strategies.

Risk management for short-term strategies tends to happen through position trading rather than portfolio construction. Assets are not held long, and portfolio characteristics change quickly. The largest risk in short-term strategies is model risk: the deployed trading strategy has already failed. Because even the best trading strategies go through periodic drawdowns, the greatest challenge for a short-term model-driven trader is deciding whether the model is in a normal drawdown or has failed completely.

Long-term model-driven investment strategies face a different set of problems. Because assets are held longer, execution cost (still important) is not the main concern. Statistical inference becomes harder, and the danger of overfitting or data-mining is greater. Risk management for long-term strategies happens at the portfolio-construction stage: because rebalancing is less frequent, you must be more careful that the portfolio is not exposed to unexpected sources of risk.

Because long-term strategies often have lower Sharpe ratios, they face a different kind of capacity problem—the manager’s tolerance for pain. There is one advantage: because you trade less often, you can have a much better lifestyle than if you run a short-term strategy.

Conclusion

I wrote this article to provide some useful information (and occasional entertainment) without telling you anything my competitors or potential competitors would find useful. Unfortunately, even knowing that it is possible to beat the market consistently may increase competition and make this kind of trading harder. So why write it? Well, one of the editors is a friend, and he asked politely. And you probably will not believe everything I tell you. If you do—well, I have always liked a challenge.

Further reading

● Grinold R C and Kahn R N, 1999, Active Portfolio Management: A Quantitative Approach for Producing Superior Returns and Controlling Risk, Probus

● Goetzmann W N, Ibbotson R G and Brown S J, “Offshore Hedge Funds: Survival and Performance 1989–1995,” Journal of Business 72, 91–117

● “Why Hedge Funds Make Sense,” Global Equity and Derivative Markets, Morgan Stanley Dean Witter, November 2000

Peter Muller is a Managing Director in the New York office of Morgan Stanley Dean Witter and head of Process Driven Trading.