Quant Analytics: from scratch to pro · Quant foundations: returns, risk and clean data · Beginner
Quant analytics: what it is and how this curriculum works
Quant analytics means asking a precise question of market data and judging the answer with care. This lesson maps the nine tracks of the curriculum and teaches the habits that run through all of them: define the question, know the base rate, count your tries, and respect what numbers cannot say.
In this lesson
- Describe what quant analytics is and name the nine tracks of this curriculum
- State a market question in a form that data can answer
- Use a base rate to judge how informative a signal is
- Explain why the best of many tries tends to look better than it is
Quant analytics is the habit of turning a question about markets into a measurement, and then asking how much the measurement can be trusted. It is not a way to find certainty. Prices are noisy, history is short, and many apparent patterns are luck. The tools in this curriculum exist mainly to tell a real effect from a lucky one.
The nine tracks
The curriculum has four levels: 101 Beginner, 201 Intermediate, 301 Advanced and 401 Pro. Nine tracks run through them.
| Track | The question it answers |
|---|---|
| Foundations | How do prices become returns, volatility and clean data? |
| Statistics | How sure can we be of an average, a difference or a beta? |
| Indicators | What does an indicator compute, and does it carry information? |
| Backtesting | What does a historical test show, and how does it mislead? |
| Portfolio | How are risk, return, sizing and diversification measured? |
| Models | What do factors, volatility models and regimes explain? |
| Microstructure | What do trading costs, depth and market impact look like? |
| Machine learning | What can learning methods add to noisy market data? |
| Practice | How is a claim read, tested, monitored and written down? |
Start with Foundations. Simple and log returns comes next, and later tracks assume it. The first lessons of other tracks are moving averages, what a backtest is, Sharpe and Sortino ratios and reading a research claim.
Start with a question that can be wrong
A usable question names the data, the event, the horizon and the statistic. "Is this market good?" cannot be wrong. "After a close below its 50-day average, is the median 5-day return lower than after other days?" can be. Write the question down before looking at results. A question chosen after seeing the data is a description of the data, not a test.
Base rates
A base rate is how often something happens in general, before any signal. It is the first number to find, because a signal only matters relative to it.
Worked example
Suppose a large one-day fall happens on 5% of days. A detector flags 80% of those days. It also flags 10% of all other days.
- Flagged event days: 0.80 x 5% = 4.0% of all days.
- Flagged ordinary days: 0.10 x 95% = 9.5% of all days.
- Share of flags that are real: 4.0 / (4.0 + 9.5) = 29.6%.
The detector catches most events, yet seven flags in ten are false alarms. The rarer the event, the more this happens. Always ask how often the thing occurs, then how often the signal fires without it.
Counting your tries
If one test has a 5% chance of a false positive, 20 independent tests have a 1 - 0.95^20 chance of at least one, which is 64.15%. Run enough ideas and one will look good. Honest work counts every try, not only the winner. The backtesting and statistics tracks return to this at length. Published research faces the same pressure, as discussed in Ioannidis (2005) and in Harvey's 2017 presidential address.
What numbers cannot say
A statistic summarises the past sample. It does not say why something happened, and it does not carry over to a changed market. Volatility, for example, describes the size of past moves; it is not a limit on future ones. Keep three questions at hand: how much data is there, how was it chosen, and what would change the conclusion?
Common mistakes
- Looking at the data first and writing the question afterwards.
- Judging a signal by its hit rate without the base rate.
- Reporting the best of many tries as if it were the only one.
- Treating a sample statistic as a fact about the market.
- Skipping the foundations because a later lesson looks more interesting.
In practice
Keep a simple log: the question, the data, the date, the number of variants tried, and the result. It takes minutes and makes later honesty possible. Analytics on FinUties applies measures like these to the markets it covers, so they are at hand when you read a chart.
Chance that at least one of N tests looks significant by luck
Worked numbers
Each result is recomputed from the formula shown.
| What | Formula | Result |
|---|---|---|
| share of days with the event | 0.05 | 5% |
| flagged event days as a share of all days | 0.80 * base | 4.0% |
| flagged ordinary days as a share of all days | 0.10 * (1 - base) | 9.5% |
| share of flags that are real | hit / (hit + false_flag) | 29.6% |
| chance that at least one of 20 null tests looks significant | 1 - 0.95**20 | 64.15% |
Check yourself
A rare event happens on 5% of days. A detector flags 80% of those days and also flags 10% of all other days. A day is flagged. How likely is the event?
Flagged event days are 0.80 x 5% = 4.0% of all days. Flagged ordinary days are 0.10 x 95% = 9.5% of all days. The share of flags that are real is 4.0 / (4.0 + 9.5), which is 29.6%. A detector that looks good still produces mostly false alarms when the event is rare.
You test 20 unrelated ideas, each with a 5% chance of looking significant by luck. What is the chance that at least one looks significant?
1 - 0.95^20 = 64.15%. Testing many ideas and reporting the winner makes a lucky result more likely than not.
Which question is testable: 'Is this market good?' or 'After a close below its 50-day average, is the median 5-day return lower than after other days?'
The second. It names the data, the event, the horizon and the statistic. The first cannot be wrong, so it cannot be tested.
Why do the tracks start with returns and volatility rather than with trading ideas?
Every later tool is built on returns, and every test needs a measure of noise. Errors in the basics, such as averaging simple returns over time, carry through all the later work.
Key terms
Sources
- Why Most Published Research Findings Are False. PLoS Medicine 2(8), e124 (John P. A. Ioannidis), 2005. Accessed 9 Nov 2026
- Presidential Address: The Scientific Outlook in Financial Economics. The Journal of Finance 72(4), 1399-1440 (Campbell R. Harvey), 2017. Accessed 9 Nov 2026