← All tools · Guides

Quantitative trading for beginners: what to learn, in what order

What to learn, what “done” means, and which page to read for each step.

This page is a separate English study plan. The English edition of the Chinese learning path is quantitative trading for beginners. This page is not that language pair. China A-share broker thresholds, fees, and the exchange filing note stay on the Chinese page. This site has no relationship with the publishers named below. There are no purchase links. Book details were checked on 22 Oct 2026 against author pages, the publisher, and Douban for Chinese editions. Anything that check did not settle is left out.

What quant trading is, and what it does not promise

Quantitative trading writes “what to buy, when, and how much” as a rule, tests that rule on history, and only then lets a program follow it. The test can show whether the rule was applied consistently. It cannot show that the rule will make money.

• Difficulty: a first backtest is a small Python job. The work after that is checking the data, accidental lookahead, costs, and whether another period still agrees.
• Profit: not guaranteed. A profitable backtest and a losing live account is a common pair, not a rare accident. Overfitting, missing fees and slippage, and a changed market are the usual reasons.
• Why learn it anyway: even if you never send a live order, the loop turns a vague idea into a rule you can check against history.

What you need before step 1

• Python: variables, lists, dictionaries, loops, functions, and the ability to read an example.
• pandas: price data is a table. You will read it, filter it, align it by date, and take rolling averages. The first book below is about that, not about trading.
• A little statistics: a mean, a standard deviation, a correlation. You need them when you read a result. You do not need a course before you start.

A local Python and SQLite setup for a China A-share review stack is the Chinese environment guide. Installing one data library and one backtester, and nothing else yet, is enough. A short list of projects by job is the open-source tool list. If you do not want to code yet, hosted China platforms are in how to choose a quant trading platform. Using AI only as a research assistant is how to choose AI stock-picking tools. Charting, a screener, and alerts, without writing a backtest, are the TradingView setup. The same walkthrough in Chinese is TradingView setup (Chinese).

Six steps

Read a result before you trust it

A good-looking backtest is the step people skip. The backtrader walkthrough names four failures. Three of them apply to any market. The fourth is about how prices are adjusted, which matters as soon as a series contains splits or dividends. The same article in Chinese is the Chinese edition.

• Lookahead: a signal computed from today's close and then filled at today's close used information the trade did not have. The backtrader walkthrough keeps the fill on the next open.
• Costs left out: a backtest with no commission and no slippage can show a profit that fees would have removed, especially if it trades often.
• Overfitting: tuning a parameter until the history looks best is fitting noise. Hold out a period that is not used for tuning, and look at it once, at the end.
• Adjustment: unadjusted prices invent jumps on split and dividend days. The backtrader walkthrough uses forward-adjusted prices (qfq) for that reason. Match the adjustment to the series you claim to have tested, and do not mix adjusted and unadjusted rows.

Return is not the only number. Maximum drawdown is the loss from a peak, and it is the number that decides whether you would have stayed with the rule. Sharpe puts return next to how bumpy the path was. The backtrader walkthrough also compares the result with a benchmark over the same dates. For China A-shares that benchmark write-up uses the CSI 300. This page does not add a numeric cutoff for a “good” Sharpe or a “tolerable” drawdown.

Paper trading: the bar before any live order

A backtest fills at a price you assumed, can hide a lookahead bug, and does not include you changing your mind. Paper trading is the same rule on data that is still arriving, with no real order. The worked version is paper trading after a backtest: a small SQLite ledger and the same fill rules as the backtest. The same article in Chinese is the Chinese paper-trading guide. The fee numbers stay on that guide. This page does not restate stamp tax, commission, or lot size. The daily paper job needs a machine that stays on. That setup is the cloud quant desk, a different article from this path and from the paper-trading ledger.

The pass bar on that guide, which this path uses as well:

• At least 10 trading days, and those days include a decline. A rising-only sample does not count.
• The gap versus the backtest over the same dates can be explained, for example by slippage or fees. An unexplained hole means stop.
• No unexplained missing or extra signal. The Chinese guide treats that mismatch as a reason to look for lookahead.
• The drawdown is one you could stand if the money were real.

Do not change parameters while the paper period is running. Write the problem down and change the rule on the next round. Passing the bar is not a reason to size up. It is a reason to consider a small live amount you can afford to lose, and to keep the log going.

What it costs to start

• Research and a backtest: can cost nothing. yfinance, AKShare, backtrader, and Qlib are free and open source and run on your own computer. backtrader is still a reasonable way to learn the ideas. For a project you expect to keep, read why its repository has gone quiet.
• Data: free libraries that read public pages can break without notice. A paid API is for when that is not good enough. Price tiers for Tushare and AKShare are on how to choose between Tushare and AKShare. The same comparison in Chinese is Tushare and AKShare (Chinese). What the English comparison has already checked for EODHD's free plan is on the data-source guide. This page does not restate those prices.
• A server: optional. You need one only if a script should run when your own computer is off. The worked version, a lightweight machine, cron, and one model API, is the cloud quant desk. The same article in Chinese is the cloud quant desk (Chinese). Opening a machine and connecting over SSH is a different article: opening a cloud server. The same note in Chinese is cloud server setup. This page does not restate any vendor discount.
• A live account: broker terminals and exchange rules differ. For China A-shares, start with how to choose a quant trading platform and how to choose between QMT and PTrade. The same deep dive in Chinese is QMT and PTrade. Thresholds and fees are per the broker's own disclosures.

After the six steps

• Machine-learning stock selection: the worked version is the Qlib walkthrough. It records a run on Windows 11 on 21 Oct 2026, after the official dataset had been paused, using the community investment_data files. The same walkthrough in Chinese is Qlib (Chinese). This page does not repeat that setup.
• A research agent: only after the data and backtest loop works. See TradingAgents, TradingAgents-CN, and Vibe-Trading. All three are research tools.
• A bot or an exchange agent: different rules, and real orders. How to choose a trading bot covers Freqtrade, OctoBot, Jesse, and the others. The same comparison in Chinese is how to choose a trading bot. For OKX, demo keys and read-only mode come before any live permission, and the key must not have withdrawal permission.

Books

Only books whose author and edition the Chinese guide checked. Chinese-edition details come from Douban listings as of 22 Oct 2026. Confirm those with the publisher. No purchase links.

• Read this first: Wes McKinney, the author of pandas, Python for Data Analysis, 3rd edition. Published August 2022, updated for pandas 2.0 and Python 3.10. The author site hosts a free online edition. The Chinese edition is China Machine Press, October 2023. The Chinese title is on the Chinese learning path. It does not teach trading. It teaches the tables you will use every day.
• After a first backtest: Stefan Jansen, Machine Learning for Trading, currently the 3rd edition. The companion code is public on the author's GitHub repository under the MIT licence, and it is listed on this site at Machine Learning for Trading. The 2nd-edition code is on the repository's second-edition branch. The Chinese guide did not confirm a publisher or year for the 3rd edition, and did not find a Chinese edition, so neither is stated here.
• When you want the machine-learning failure modes: Marcos López de Prado, Advances in Financial Machine Learning, Wiley, 2018. It is about the mistakes that are easy to make with machine learning on financial data. It is math-heavy, and it is a poor first book. The Chinese edition is CITIC Press Group, 2021. The Chinese title is on the Chinese learning path.
• Optional: Yves Hilpisch, Python for Finance, 2nd edition, O'Reilly. It covers a lot of ground. The author's code repository says the code is based on Python 3.7, so typing along means watching for version gaps. The Chinese guide did not find a Chinese translation of this edition.

This is not investment advice. Quant trading can lose money. A backtest and a paper account are not a forecast. Books, platforms, and brokers follow their own current notices.