Derivatives and advanced market conceptsModule 5, lesson 21 of 21
Quantitative analysis
Turning a hypothesis into a tested model, factors such as value and momentum, data pitfalls, and why simple models and unpublished edges hold up better.
In this lesson
- Describe the workflow from hypothesis to tested model.
- Explain common factor ideas such as value, momentum, quality and low volatility.
- Identify data problems that make results look better than they are.
- Explain why complex models often fail out of sample and why published anomalies tend to weaken.
Builds on
- Overfitting, sample size and out-of-sample testingOverfitting is the central danger in quantitative work.
- Correlation and causationQuantitative models find correlations that may not be causes.
On this page
In simple terms
Quantitative analysis uses data and statistics to make investment decisions by rules rather than by judgement case by case. A quant turns an idea, such as "cheap companies tend to do better", into precise rules, tests them on historical data, and checks whether the results hold on data the model has not seen.
The method is powerful, and it is also easy to fool yourself with it.
The workflow
- Hypothesis: a reason why something should predict returns, such as investors overreacting to bad news.
- Data: prices, financial statements and other inputs, cleaned and checked.
- Rules: precise definitions of what to buy and sell, and when.
- Backtest: apply the rules to past data, including realistic costs, as the backtesting lesson explains.
- Out-of-sample test: check on data not used to build or tune the model, as the overfitting lesson explains.
- Small live trial and monitoring: real markets add costs and frictions that tests miss.
Factor ideas
Research over decades has linked certain characteristics, called factors, with differences in returns:
| Factor | Idea |
|---|---|
| Value | Shares that are cheap relative to earnings, book value or cash flow |
| Momentum | Shares that have risen strongly over the past several months |
| Quality | Companies with high, stable profitability and strong balance sheets |
| Size | Smaller companies |
| Low volatility | Shares whose prices have swung less |
Explanations differ: some see factor returns as payment for extra risk, others as the result of investor behaviour. Each factor has gone through long periods, sometimes many years, of underperformance, and results vary between markets and periods.
Data problems
- Survivorship bias: testing only on companies that still exist ignores the ones that failed, as the backtesting biases lesson explains.
- Look-ahead bias: using information that was not available at the time, such as annual results before they were published.
- Ignoring costs and impact: strategies that trade small, illiquid shares often look good until costs are included, as the microstructure lesson explains.
- Data errors: wrong prices, missing corporate actions, unadjusted splits.
Simple versus complex
Worked example: Two made-up models
| Model A | Model B | |
|---|---|---|
| Settings tuned on past data | 15 | 2 |
| Economic reason | None stated | Clear |
| Backtest return per unit of risk, in sample | Excellent | Moderate |
| Same measure, out of sample | Close to zero | Moderate |
With many settings, Model A fitted the noise in its data. Model B, simpler and grounded in a reason, held up. This pattern is common in quantitative research.
The more settings tried and tuned, the more likely a model is to find patterns that were chance, as the correlation lesson warns.
Anomalies fade
When researchers publish a pattern that predicted returns, investors start trading on it. Studies of many published anomalies have found that their returns tend to be noticeably lower after publication than in the original research. An edge that everyone knows about is harder to profit from.
Why it matters
Quantitative claims, from factor funds to trading algorithms, are increasingly marketed to individuals. Understanding how models are built and tested, and where they fail, helps you judge such claims, and shows why simple, well-reasoned, honestly tested ideas deserve more trust than impressive backtests.
Limitations
This lesson introduces ideas in general terms; quantitative research involves statistical methods well beyond its scope. Factor evidence varies across markets, periods and definitions, and past research does not guarantee future results.
Common mistakes
- Trusting in-sample results.
- Ignoring costs, survivorship and look-ahead bias.
- Preferring complexity over a clear rationale.
- Expecting factors to work every year.
Try it yourself
- Write a simple hypothesis about share returns and explain why it might work.
- Turn it into precise rules a computer could follow.
- List the data you would need, and the biases that could creep in.
- Decide how you would split the data between building the model and testing it.
Key takeaways
- Quantitative analysis turns an idea into precise rules, tests them on data, and checks the results on data not used to build them.
- Factor ideas, such as value, momentum, quality, size and low volatility, are characteristics that research has linked to differences in returns over long periods. Each can underperform for years.
- Data problems, such as survivorship bias, look-ahead bias and ignoring costs, can make a model look far better than it is.
- Complex models with many tuned settings often fit past noise and fail on new data. Simple models with a clear reason to work tend to hold up better.
- Research has found that the returns of published anomalies tend to shrink after publication, as investors trade on them.
Check your understanding
Finished reading?
Your progress is saved only in this browser.
Terms used in this lesson
- Quantitative analysis
- Making investment decisions with data, statistics and precise rules, tested on past data and checked on data not used to build them.
- Factor investing
- Investing based on characteristics, such as value, momentum, quality, size or low volatility, that research has linked to differences in returns.
- Overfitting
- Tuning rules so closely to past data that they capture random noise rather than a lasting pattern, so they look excellent in tests and fail afterwards.
Spotted a mistake or something unclear? Report it by email. Here is how corrections work.
This lesson is general education, not investment advice. Examples use made-up or clearly dated figures to explain ideas; they are not recommendations. Read the disclaimer.