Build a base in probability, statistics, linear algebra, and Python, then complete a research project with an explicit hypothesis, a benchmark, and a chronological evaluation. Add role-specific material after that. A systematic equities researcher may need factor models and panel data, an options researcher may need stochastic calculus and numerical pricing, and a market microstructure researcher may need order-book data and execution models.
Build evidence you can show someone
- Build the core. Practise conditional expectation, variance, regression, sampling uncertainty, and matrix algebra. Use Python to load and inspect time-stamped data, reproduce an ordinary least-squares estimate, and compare the result with a standard library. Write a short explanation of what the estimate means.
- Learn the market mechanics your project needs. For equities, understand returns, dividends, splits, portfolio weights, and trading costs. For derivatives, start with payoffs, no-arbitrage, replication, and a one-period binomial tree. Learn the data definitions before fitting anything.
- Run one complete experiment. Write the hypothesis and evaluation rule first. Compare against an obvious baseline, preserve the chronological split, document assumptions, and show results under more than one cost scenario if the experiment implies trading.
- Choose a research direction. Read a paper or lecture in your target area. Reproduce one result with an accessible dataset, then change one assumption and explain what happens. Prepare to discuss the experiment, your errors, and what additional evidence would change your mind.
Move on when you can complete each phase's deliverable without leaning on a tutorial for every step. A fixed number of weeks matters less than those checkpoints.
A first project you can finish
A first project could ask whether recent realised volatility improves a one-day-ahead variance forecast for an equity index. Define the target as the squared return at and make the forecast from data available through . Compare a 20-day average of squared returns with a longer expanding-window average. Choose an error measure ahead of time and evaluate on a later sample.
If the rolling forecast improves results in turbulent periods and performs worse in calm periods, report both. Explain that a daily squared return is a noisy volatility proxy. Your write-up can also show whether the conclusion changes across time windows. This scope is small enough to finish and rich enough to discuss in an interview.
A useful finish line is a two-page research note and a script that reproduces its main figure. Explain when each input becomes available, why you chose the benchmark, and what happened on the untouched dates. Keep the negative results. If the simple expanding forecast wins, work out whether the rolling estimator was noisier, whether the result depended on a crisis, or whether neither difference was large relative to sampling noise. That investigation is closer to quant research than turning a weak result into a polished dashboard.