Strategy family 03 / 12
Factor investing & quantitative selection
Factor investing selects securities systematically, using characteristics that can be measured across a whole market, such as value, quality, size, low volatility and momentum. Multi-factor models combine several such signals into a single forecast, and smart beta packages factor exposures as transparent, rules-based indexes. Index enhancement aims to beat a specific benchmark within a tight tracking-error budget, while quantamental work turns fundamental judgment into signals that can be tested at scale.
Covers
- Multi-factor
- Smart beta
- Index enhancement
- Quantamental
Why it is a demanding test
Factor research is a test of statistical discipline. Signals are weak relative to noise and the number of possible variations is enormous, so a backtest can look good almost by accident; the work rewards systems that state hypotheses in advance, correct for multiple testing and budget for costs and capacity. It also demands exact handling of data timing, since using data even a few days before it was published can manufacture returns that never existed.
What we test
What a system needs to do well in this family. Tasks are built so that each ability can be observed and scored on its own.
Computing signals without look-ahead
Calculates factor values only from data available on each date, with realistic reporting lags, and handles outliers, missing values, corporate actions and delistings. Neutralizes industry or size exposure where the design calls for it.
Encoding fundamental ideas as signals
Turns an analyst's idea, such as earnings quality or pricing power, into a precise, reproducible definition. Checks whether it adds information beyond well-known factors rather than repackaging them.
Testing signals with statistical care
Measures predictive power with rank information coefficients, quantile spreads and genuinely out-of-sample periods. Accounts for how many variants were tried and checks that results hold across sectors, size groups and time.
Combining signals into one forecast
Weights signals with regard to their correlations, decay speeds and turnover, and explains what diversification the blend provides over its strongest single component.
Constructing portfolios within a risk budget
Turns forecasts into holdings through transparent weighting rules or an optimizer with a risk model, limiting tracking error, sector and style exposure, turnover and position size. Attributes results to factor exposures, stock-specific returns and costs, so an intended edge can be told apart from an accidental style bet.
Example task types
Illustrative designs that show the shape of the work. They are not items from the benchmark itself.
- Track: Analysis
Audit a factor backtest for hidden flaws
The system receives the code, data description and results of a multi-factor backtest with several planted flaws, such as fundamentals used before their publication date or a universe built from today's index members. It must find and explain each flaw, estimate its likely effect on the results and propose a fix. A good answer ranks the flaws by impact and does not report problems that are not there.
- Track: Simulation
Run an index-enhancement mandate
In a historical replay, the system manages a portfolio against a broad equity index with a stated tracking-error budget, sector and style limits and a turnover cap, rebalancing on a fixed schedule with only the information available on each date. Trades pay commissions and market impact scaled to each stock's liquidity. Scoring looks at excess return per unit of tracking error and at whether every constraint held, including when the budget came under pressure.
- Track: Live
Rank a stock universe every week
Each week, the system submits a ranked forecast of next week's returns for every stock in a fixed universe, along with the definitions of the signals it used. Forecasts are scored by rank information coefficient and its stability over a rolling window, and the implied portfolio is tracked after estimated costs. Consistency across weeks counts for more than a few strong periods.
- Track: Client
Compare smart beta with a plain index
A user comparing a low-volatility smart beta fund with a market-cap-weighted index fund asks which one is better. The system must explain how each is built, what factor exposures and sector tilts the smart beta fund carries, how it may behave in a sharp market rally, and how fees and turnover differ. A good answer does not present historical factor premiums as guaranteed and ties the comparison back to the user's goals.
How it is scored
Each run is scored on its outcome and on the process behind it, so a lucky result is not mistaken for a sound one.
- 01
Trial-adjusted information ratio
Active return per unit of tracking error against the mandate's benchmark. In research tasks it is deflated for the number of variants tried, so the best result of a wide search is not rewarded as if it were the only attempt.
- 02
Rank information coefficient
The rank correlation between forecasts and subsequent returns across the universe, and how stable it is from one period to the next.
- 03
Net return and capacity
Performance after commissions, spreads and market impact, with an estimate of how quickly it erodes as assets grow. A signal that looks strong before costs may not survive this step.
- 04
Exposure and constraint adherence
How closely realized factor, sector and stock exposures match what the strategy says it targets, and how often stated limits are breached.
- 05
Research hygiene
A rubric-based review of whether hypotheses were stated before testing, how many variants were tried and disclosed, whether out-of-sample periods stayed untouched, and whether data timing was handled correctly.
Diagnostics
Failure modes we diagnose
Recurring mistakes the tasks are built to surface. Each one is reported separately, so builders can see which to fix first.
- 01
Overfitting through repeated search
Tunes parameters, universes and holding periods until the history looks good, so the result reflects the search more than the signal. Undisclosed variants are the usual symptom.
- 02
Leaking the future into the past
Uses fundamentals before their publication date, revised figures in place of the originals, or a universe made only of today's survivors. Each makes a backtest look better than anything that could actually have been traded.
- 03
Ignoring turnover, costs and capacity
Reports gross returns from a fast-decaying signal without charging for the trading it requires, or assumes a small-cap strategy can absorb far more capital than its stocks can trade.
- 04
Confusing style tilts with skill
Presents excess return that really comes from an unintended tilt, such as toward small caps or a single sector, as stock-selection skill. In index enhancement this often stays hidden until the tilt reverses.
- 05
Overlooking crowding and correlated exits
Treats a popular factor as safe without considering that many portfolios hold similar positions and may need to sell at the same time.
Environment and data
Factor tasks use point-in-time fundamentals stamped with publication dates, historical index membership and a survivorship-free universe that includes delisted and merged companies, with prices adjusted for corporate actions. Simulations charge commissions and spreads, plus market impact that grows with trade size relative to daily volume, and apply practical limits such as short-sale availability, trading suspensions and daily price limits where a market has them. We hold back a final out-of-sample period that systems never see during development.