Quantitative Trading

Systematic Strategies

Thousands of small, diversified positions in U.S. equities and liquid futures, each chosen and sized by models that had to pass the same tests before they traded.

Objective

The strategies seek returns that are largely independent of the direction of stock and bond markets, with low correlation to the S&P 500 over full market cycles. They are intended as a diversifying allocation for institutions and qualified investors, not as a replacement for core holdings.

Each strategy earns many small, uncertain edges at once. No single position is expected to matter much; the result depends on the number of independent bets, the accuracy of the models and the cost of trading them.

Approach

  • Signals. Forecasts of each security’s return over the next days to weeks, from prices, company fundamentals, analyst revisions, text from filings and news, and flows.
  • Combination. Signals are blended with weights estimated on past data and shrunk toward equal weights, which tends to hold up better out of sample.
  • Construction. An optimizer turns forecasts into positions that maximize expected return after trading costs, subject to neutrality and risk limits.
The question is never whether a backtest looks good. It is how many backtests were run to find it.
Best Sharpe ratio expected by chance, by number of strategies tried Bar chart. With ten years of data and no real skill, the best of 10 random strategies is expected to show an annualized Sharpe ratio of about 0.5; the best of 100, about 0.8; the best of 1,000, about 1.0; and the best of 10,000, about 1.2. 00.51.0 101001,00010,000 Number of strategies tried Best annualized Sharpe ratio, no skill 0.490.791.021.22
Expected maximum of independent, zero-skill strategies over 10 years of data; a simplified illustration of the multiple-testing problem, not Lenox results.

Strategies

Three families of models.

Holding period: 5 to 20 days

U.S. Equity Market Neutral

Long and short positions in about 2,000 liquid U.S. stocks, balanced so that market beta, sector and style exposures are close to zero. Returns come from relative moves between companies.

Holding period: hours to days

Short-Horizon Futures

Positions in about 40 of the most liquid futures contracts, based on intraday price patterns, order flow and the lead-lag between related markets.

Combines both

Multi-Strategy

A single portfolio of the equity and futures models, with risk allocated so that no one model drives results and capital moves to where costs are lowest.

An engineer reviewing code on a large monitor

Research standards

Luck looks like skill if you try enough.

Test a thousand strategies with no real edge on ten years of data and the best one will usually show a Sharpe ratio near 1.0. That is why our researchers log every trial, and why the hurdle for a new signal rises with the number of trials behind it.

  • Pre-registered hypotheses

    The idea, the data and the test are written down before results are seen.

  • Point-in-time data

    Every input is used only as it was known on the date in question, including restated earnings and index changes.

  • Untouched hold-out period

    Recent data stays sealed until the design is final, and is opened once.

  • Independent replication

    A second researcher rebuilds the result from the written specification before it can trade.

Read our paper on out-of-sample testing

Portfolio and risk

Neutral by construction, limited by design.

Risk management at Lenox

Neutrality constraints

Equity portfolios are held near zero market beta and within tight limits on each sector and on common style factors such as size, value and momentum, measured with a commercial and an internal risk model.

Capacity

Trading costs rise with size. Each strategy has an estimated capacity, set where expected costs would consume a third of expected return, and it closes to new capital before reaching it.

Limits and drawdown rules

Gross exposure, single-name size, factor exposures and daily loss are limited by the risk team. A drawdown beyond a preset level triggers an automatic cut in exposure and a formal review.

Crowding

Popular quantitative signals can unwind together, as they did in August 2007. We monitor overlap with widely known factors and limit the share of risk they may take.

Characteristics

Typical terms.

Typical strategy characteristics
CharacteristicTypical range
Target annualized volatility6% – 10%, depending on strategy
Market beta (equity strategies)−0.05 to +0.05
Gross exposure (equity strategies)4 to 6 times net assets
Average holding periodHours to 20 trading days
Fund liquidityMonthly, with 30 days’ notice
Fund eligibilityQualified purchasers; minimum $5 million
Separate account minimum$100 million

Risks

Systematic strategies can lose money, including in periods when markets rise. Models are built on historical relationships that may weaken or reverse. Market-neutral portfolios use leverage and short selling: short positions can lose more than their initial value, and borrowing costs or recalls can force sales. Signals shared with other managers can unwind at the same time. Technology and data failures can cause unintended trades. No performance is shown on this page, and hypothetical results have inherent limitations.

Fund interests are offered only by private placement memorandum to eligible investors. The S&P 500 is named for context only; Lenox Asset Management has no relationship with any index provider.