Quantitative analysis is the evaluation of securities, companies, portfolios or markets through measurable data and mathematical and statistical methods. In investing it covers everything from computing ratios out of a company's financial statements to estimating risk with probability models and building automated trading systems. The defining feature is that conclusions rest on numbers that can be measured, tested and repeated rather than on personal judgement alone.
It is usually contrasted with qualitative analysis, which weighs factors such as management quality and brand strength, and the two together make up fundamental analysis of a company. Specialists who build and run such models are known as quants, and the wider academic field is quantitative finance. The same methods are used by banks, asset managers, insurers, regulators and corporate finance teams.
Main techniques
The toolkit ranges from simple arithmetic to advanced statistics:
- Ratio analysis. Measures such as margins, return on equity, leverage and valuation multiples are computed and compared over time and against peers.
- Valuation models. A discounted cash flow model estimates what future cash flows are worth today, while option pricing models value contingent claims.
- Statistical estimation. Regression analysis and time series analysis measure relationships between variables, such as how a share responds to the market or to interest rates.
- Factor models. A factor model explains returns by exposure to common drivers such as value, size and momentum.
- Simulation and risk measures. Monte Carlo simulation generates thousands of possible outcomes, and measures such as value at risk summarise the potential for loss.
In recent decades machine learning methods and alternative data, such as satellite images or card spending records, have been added to these traditional tools.
Example: estimating beta
A common quantitative task is to measure how sensitive a share is to movements in the overall market, its beta. It is estimated from historical returns as:
- is the return on the share over each period, such as a month.
- is the return on a market index over the same periods.
- is the covariance of the two series and the variance of the market return.
Suppose five years of monthly data give a covariance between the share and the index of 0.0018 and a variance of the index of 0.0012. Beta is , so the share has tended to move about 1.5 percent for each 1 percent move in the market. The figure can then feed the capital asset pricing model to estimate a required return, or be used to size a hedge.
Applications
Quantitative analysis is used to screen thousands of companies for those that meet set criteria, to construct portfolios that balance expected return against risk, and to measure performance with statistics such as the Sharpe ratio. Banks use it to price derivatives, set capital and score borrowers. Trading firms use it to design and test strategies, and quantitative trading funds run entire portfolios on model signals with little human discretion. Regulators rely on it for stress tests that project how institutions would fare in severe scenarios. Corporate treasurers use similar methods to forecast cash needs and to measure exposure to currencies and interest rates, and insurers use them to price policies and set reserves. In each case the aim is to replace rough estimates with figures that can be checked, compared and updated as new data arrive.
Analysts commonly work in statistical languages such as R and in general-purpose languages, as well as in spreadsheets for smaller tasks. Strategies are usually checked by backtesting them on historical data before any money is committed.
Limitations
Models are only as good as their data and assumptions. Historical relationships can break down when markets change, and measures estimated in calm periods may badly understate risk in a crisis, when correlations between assets tend to rise. Many financial return series have fatter tails than the normal distribution assumes, so extreme losses occur more often than simple models predict.
A second danger is overfitting: testing many variations on the same data can produce a model that fits the past perfectly but has no predictive power. Crowding is a third: when many firms run similar models, their trades can move prices against one another, as happened in the sudden losses suffered by several quantitative equity funds in August 2007. For these reasons banks maintain formal model risk management, and most investors combine quantitative results with qualitative judgement.