Algorithmic Trading by Ernest P. Chan: research reading companion
Read Chan for the transition from an idea to a causal research specification; work a residual and execution-timing exercise.
Editorial reading notes and source-based resource assessments, not a claimed cover-to-cover reviewer programme. Edition, access and evidence limits are stated.
What the checked contents establish
Wiley identifies Ernest P. Chan's Algorithmic Trading and lists backtesting, automated execution, mean reversion, momentum and risk-management material. That supports a subject-based reading decision. It does not establish that our desk read every chapter, reproduced the examples or verified the author's trading outcomes. Use the publisher record to match the actual title and format rather than assume that every Chan book is the same resource.
The practical fit is a reader who wants to specify and investigate a systematic idea. If you cannot explain a return, a lagged observation and a training sample, begin with those prerequisites. A software example can illuminate a mechanism while its environment or data access requires a separate current check.
Separate a statistical object from an order
Our proposed reading exercise begins with an invented residual r = A - 2B - 5. At A = 110 and B = 50, r equals 5. Write what information would be needed before interpreting that deviation: the development window, coefficient estimation, normalization, eligible instruments and decision time. Do not assume that a visually large residual must revert or that the book endorses the fixture's coefficient.
Next describe the intended position independently. A short one unit of A and long two of B has a particular quantity relationship; contracts, currencies and multipliers could require different quantities. A statistical coefficient cannot be copied blindly into lots. The exercise's output is a specification with unresolved fields, not a discovered pair or an observed profit.
Write the earliest executable clock
Suppose an idea uses a completed daily close. Record when that close becomes available, when the proposed rule can be calculated and the first eligible later order. A backtest that buys at the already completed close needs a justification beyond knowing that number later. Use a deterministic timing fixture to expose any hidden same-bar assumption before interpreting market results.
Keep source revisions and historical membership in the same clock record. Today's adjusted data or surviving universe may differ from what existed at an earlier decision. A successful port of an example into another language shows software behavior under your assumptions. It does not independently reproduce the original research or prove that the new implementation has an economic edge.
Build a small experiment before a large grid
Write one baseline, one candidate and a rejection criterion before changing thresholds. Keep entry, exit, holding limit, sizing and costs in separate fields. Record every outcome-informed revision with a parent candidate ID. A notebook that keeps only the best curve leaves the reader unable to reconstruct the selection that produced it.
Start with arithmetic and state-transition fixtures: unavailable data, duplicate observations, a gap through an intended exit and partial quantity. Then identify a genuinely reserved evaluation design. Correctness tests and later market observations have different jobs. Counting synthetic fixtures as validation trades would make the evidence look larger without adding independent information about strategy behavior.
Check prerequisites against the actual task
Use this title as a strategy-research reading candidate when the missing skill is translating ideas into causal calculations. Use Quantitative Trading instead when the immediate gap is the workflow and economics of running a research operation; the titles overlap but their stated emphasis differs. Neither publisher description certifies a production deployment or account suitability.
Choose one sample or contents item and try writing its input-output contract in plain language. If you cannot identify what is known before the outcome, pause implementation and address that prerequisite. Buying more data or a larger cloud instance will not resolve an undefined target. Keep any purchase decision separate from an expectation that the book pays for itself.
The reading output and its evidence boundary
Complete the worksheet with the source location you actually inspect, your paraphrase, your own operational interpretation and the unanswered execution questions. An attributed rule needs an identifiable edition and passage; a threshold you invented belongs in your interpretation column. Preserve failed attempts rather than rewriting the notes to imply that the first implementation worked.
This companion is based on the publisher's bibliographic record and contents, not a documented full-book review. No star score, reader survey or replicated return is supplied. The useful endpoint is a small, inspectable experiment design. A later negative result, missing source record or rejected execution assumption can be a valid learning output without being presented as evidence that every idea in the book fails.
Reading decision checkpoints
| Question | Useful output | Do not infer |
|---|---|---|
| Read for | Causal strategy specifications | A ready-made profitable account |
| Produce | One frozen experiment and timing fixture | A large unrecorded parameter search |
| Compare with | Quantitative Trading for operational workflow | All Chan titles treated as interchangeable |
Download blank reading record (CSV)
Headers only. No completed reading sessions or reader outcomes are supplied.
Sources and scope
Choose a book by the skill you need
Keep the selected format beside the causal specification
The new register adds the specific hardcover ISBN 9781118460146 and its publisher-listed 2013 format date. Its contents and catalogue scope were checked, while the newly linked sample did not load successfully. That gap does not erase the existing useful companion or become a full-book review. A reader should attach the actual inspected location to any implementation claim. A causal fixture still tests software rather than an independent market sample, and book-inspired instrument work must complete the canonical full factory and independent master.