A survival library — what to read for a specific problem

Start choosing your next reading material by asking what you need to learn. Different resources help explain order execution, organise risk measurement or assess how strategies are tested. This lesson offers selected books and other sources, with specific editions, legal access options and tasks to complete yourself. It also shows when a course, community or instructor can help, and when to postpone another purchase. The aim is to understand a chosen subject better and use it in a practical task, not simply to expand your library. A worked reading card helps separate correct arithmetic from a claim about future strategy performance.

An illustration introducing the lesson. Model results and conditions are described separately in the article.

Give the next resource a question you can test

A useful reading task ends with an inspectable result. Suppose the question is how to distinguish choosing a prediction rule from checking it on new data. The authors’ Statistical Learning website provides edition information and legal book access; check the Python edition and its errata when following that material.

Our miniature exercise below is entirely invented for the lesson. It is not quoted from the book, a currency forecast or a benchmark of any actual model. Its numbers have no market units. The task is to calculate mean absolute prediction error and identify which data were used to choose the rule.

Record the source, edition, question and your own calculation separately. A familiar formula does not establish that you understood the experiment producing a reported result. Nor does completing a toy exercise establish expertise in the whole subject.

Calculate the loss function you actually selected

The first set has observed targets 1,2 and 6. Candidate A always predicts 2; candidate B always predicts 3. Absolute error is the nonnegative distance between a prediction and its observed target. Mean absolute prediction error averages those distances, rather than allowing positive and negative errors to cancel.

Hypothetical prediction exercise
Observed targetA predicts 2: absolute errorB predicts 3: absolute error
112
201
643
Sum56

A has mean error 5/3 ≈1.67. B has 6/3 =2. Under this chosen loss function, A scores better on these three observations. This says nothing about trading returns, transaction costs, directional accuracy or the consequences of a large error.

The phrase mean absolute prediction error is used explicitly here. It is different from the journal’s maximum adverse excursion, which is another concept sometimes abbreviated with the same letters.

Freeze the selected rule before opening the next cases

Both candidates were inspected on the first set, and A was selected after that comparison. The first set therefore belongs to the selection process for this decision. Calling it an untouched test after choosing A would misdescribe the process.

In the hypothetical continuation, record the decision to keep A fixed before revealing the next two targets:2 and 5. A’s absolute errors are 0 and 3, giving mean error 3/2 =1.5. B would produce errors 1 and 2, also averaging 1.5, but looking at that comparison does not allow you to retroactively claim that B was the preselected rule.

Two invented new cases verify the procedure and arithmetic; they do not establish a generalisation estimate for a real financial model. If you start adjusting the rule using them, record that use and obtain a genuinely separate check for the revised choice.

State what the reading changed and what remains unknown

A supported conclusion is: “I can reconstruct the error calculation and identify which observations were used for selection.” It is not: “This forecast makes money.” The exercise has no executable prices, positions, timing, financing or market data, so there is no financial result to calculate.

The next learning step might be reproducing the calculation from a small file, trying a different loss function or explaining why cancellation of signed errors can be misleading. Change one question at a time and retain the earlier work. Check any published correction against the edition actually used, rather than silently combining different versions.

The full lesson below retains its source comparisons, legal access links and original critical-reading card. Use those resources to select material for a demonstrated gap. A valid outcome can be a better question and a decision to continue studying without changing a strategy or taking a position.

Three exercises with worked answers

The exercise data are hypothetical. Choose an answer, then check the calculation.

1. What is A’s mean absolute prediction error on the first set?

Targets 1,2,6; A predicts 2 each time; absolute errors 1,0,4.

Show the answer and explanation

B. State the chosen loss function and the observations over which it was averaged.

  • A. That is the total, before averaging.
  • B. Correct:(1 +0 +4)/3 =5/3, approximately 1.67.
  • C. Absolute errors do not cancel by sign.
  • D. That is the prediction value, not the mean error.
2. Which description matches selection of A?

Both A and B were compared on the first three targets before A was chosen.

Show the answer and explanation

D. Preserve the selection history. Renaming the same data does not remove their earlier use.

  • A. The choice explicitly used those observations.
  • B. Both candidates were compared.
  • C. The exercise contains no trading result.
  • D. Correct: the observed comparison formed part of the selection process.
3. What does the continuation establish for fixed A?

New targets 2,5; A remains 2; absolute errors 0,3.

Show the answer and explanation

C. Keep the result within its small hypothetical scope. Any new tuning must be recorded as another selection step.

  • A. An error score does not establish a financial edge.
  • B. That is the total error, not its mean.
  • C. Correct:(0 +3)/2 =1.5 for the stated new cases.
  • D. The scenario explicitly freezes A before revealing the targets.

From a checked exercise to a purposeful reading plan

Continue with the original library, source-access notes and reading card. They connect a learning need to an appropriate source, a reproducible task and a limited conclusion about what you now understand.

Choosing material for a need

First name the gap: “I cannot distinguish planned risk from realised loss” is a useful starting point. “I need to read more about the market” is too broad to assess the result. Check the contents, prerequisites and a sample, then choose a chapter. You need not read every book cover to cover or buy the entire list below.

The selection covers five resources with different uses, not a ranking of “the best books for everyone”. Bibliographic details and legal access points were checked against author, publisher and institutional sources on 8 September 2026. A catalogue link confirms the edition and scope described by the publisher; it does not mean free access to the full text. The tasks below are original educational exercises, not accounts of implementation.

An older publication may still explain a concept well, but must be compared with today's market. A newer date does not guarantee a correct method. For legislation, terms, APIs and prices, use the relevant organisation's current documentation rather than a historical example in a book.

Five resources, five applications. Choose for the problem; not a compulsory canon or a ranking of all knowledge.
Choose for the problem; not a compulsory canon or a ranking of all knowledge. Check lawful access, prerequisites and current documentation where rules change.
Match the source to the question
Match the source to the questionA textbook explains a mechanism; the relevant institution supplies the current rule; a research assessment needs methods and data; a skill is practised with a task and explicit checking criteria.Understand a mechanismA textbook explainingassumptions and examples.Check a current ruleThe relevant institution andthe document applicable tothe situation.Assess a research resultMethod, data, limitations andsample selection.Practise a skillA task with data, criterionand a check solution.

The source category starts the selection; it does not guarantee quality. Check scope, edition, date, assumptions and whether the material answers your question.

Risk, microstructure and FX standards

Van K. Tharp — the language of risk and position size

Trade Your Way to Financial Freedom, 2nd edition, McGraw-Hill, ©2007, English, ISBN 9780071478717. The publisher's catalogue gives a November 2006 release date; this ISBN identifies the print edition, not the e-book. Legal access: purchase from the publisher or a library copy.[1]

Gap and level: the basics of describing results in risk units and choosing position size, after understanding percentages and trade results. The contents include chapters on R-multiples, expectancy and position sizing. Limitation: naming a risk unit does not guarantee that a loss cannot exceed 1R; execution conditions and price gaps need separate assessment.

Task: record the initial planned risk R₀ and net result for three synthetic trades, then calculate their R outcomes. Check what happens when a loss exceeds R₀ and when R₀ was not recorded. A completed example appears in the reading card. Update: compare order and cost assumptions with current instrument and account documentation.

Larry Harris — how trade execution works

Trading and Exchanges: Market Microstructure for Practitioners, Oxford University Press, 2002, English, ISBN 9780195144703. The year and ISBN follow OUP's catalogue for this edition. Legal access: purchase or library/licensed access through Oxford Academic.[2]

Gap and level: orders, liquidity, spread and participant roles, for readers familiar with basic order types. The book helps separate an investment idea from its execution conditions. Limitation: historical market structure is not a specification for today's CFD broker or a universal model of the FX market.

Task: prepare two synthetic quotes with bid/ask prices, time, size and source name. Identify what is missing to compare actual execution costs, such as commissions, available liquidity and order-handling rules. Do not call a difference between sources an achievable arbitrage. Update: supplement your market description with current execution policy and data specifications.

Global Foreign Exchange Committee — wholesale-market standards

FX Global Code, December 2024 version, GFXC, institutional material, English. The official GFXC site provides free, legal access to the document and translations. It contains 55 principles of good practice in the wholesale foreign exchange market; it does not replace law or regulation.[3]

Gap and level: ethics, information sharing and organising FX transaction work; also useful when preparing for an operations or markets role. Limitation: the document is not a retail trading manual, a professional licence or confirmation of a particular organisation's compliance.

Task: choose a communication principle and describe a fictional professional situation: who shares what information, with whom, for what purpose and on what basis. Identify the necessary control and what the Code itself does not settle. Update: before professional use, check the version on GFXC's website and your organisation's procedures.

Research and statistics — when the project requires them

David Aronson — assessing rules and the risk of overfitting

Evidence-Based Technical Analysis: Applying the Scientific Method and Statistical Inference to Trading Signals, 1st edition, Wiley, 2006, English, ISBN 9780470008744. These details identify the edition in the Wiley-VCH catalogue; legal access is through the publisher or a library.[4]

Gap and level: distinguishing a hypothesis, test result and selection of favourable variants; basic statistics and an understanding of backtesting are required. Limitation: historical examples do not establish the effectiveness of your own FX strategy, and one good result does not remove the risk of multiple testing.

Task: describe one testable rule, register every variant considered and design a chronological data split. Record which information must remain unavailable while selecting the rule. A plan alone is not a completed test. Update: actual research requires current data documentation, costs and a method for controlling dependence and selection.

James, Witten, Hastie, Tibshirani and Taylor — statistical learning basics

Gareth James, Daniela Witten, Trevor Hastie, Robert Tibshirani, Jonathan Taylor: An Introduction to Statistical Learning: with Applications in Python, 1st edition, Springer, 2023, English, print ISBN 9783031387463, DOI 10.1007/978-3-031-38747-0. The authors' website provides a legal PDF and materials; Springer's catalogue identifies the edition. This is not the edition with R labs.[5]

Gap and level: validation, prediction error and model selection; basic Python and statistics are helpful. Limitation: a general statistical learning textbook is not a ready-made financial backtesting method. Randomly splitting observations may be inappropriate for dependent market data.

Task: reproduce one lab using its data, record tool versions and compare the result with the material. Then describe what would need to change for a time series before using the method on exchange rates. These are two separate stages. Update: check the Python edition's errata and library documentation; a version difference may change the code without changing the idea discussed.

Video and podcasts as supplements

If listening helps you learn, the authors' website for An Introduction to Statistical Learning also links to courses and recordings. Match labs to the language of your chosen edition; similar lectures do not imply identical code. Access to material and a paid completion certificate may have different terms.[5]

A conversation with a practitioner may help frame a question, but a guest's recollection is not a dataset. Record the episode date, the speaker's role, any product promotion and one claim to check. If methods or data are unavailable, label it an opinion or an unverifiable account. An interesting interview need not prompt a portfolio change.

Conferences and communities — why take part

A useful purpose may be discussing a specific problem, reviewing a method or understanding a role's actual scope. Before buying a ticket, check the programme, speakers, access to materials and travel costs. In a discussion group, assess whether participants cite sources and can identify limitations of their own results. Community size and whether access is paid or free do not determine quality.

Bring a short question and a safe example to a discussion. Do not publish client data, credentials or employer material without permission. A synthetic project often allows discussion of a method without disclosing confidential information. The meeting's value may be a better question or a detected error, not a contact leading to a trade.

Choose tools for the exercise

A document or spreadsheet is enough for a reading card. A Python lab needs an environment compatible with the material. Only a task requiring particular data or features justifies assessing a paid service. Check the licence, data period and frequency, time zones, adjustments and export options; “historical FX data” alone is insufficient.

Limit purchases to what is needed now. Record data sources, download dates and code versions. If an exercise depends on a missing paid feature, you can choose other data or stop at designing the method — just describe the completed scope clearly.

When an instructor helps

A course makes sense when it solves a specific problem not covered by available material: work assessment, consultation or a structured path with prerequisites. Ask for an example of feedback and how tasks are assessed. Do not assume that price signals quality or that investment skill automatically means teaching skill.

Compare the full cost with self-study and a single project review. Show the time for each route separately; without later income data, a return cannot be calculated. A completed example is in the chapter on education costs.

When to postpone a purchase or resource

Set material aside if you lack its prerequisites, it does not answer your current question or it duplicates existing resources. Reject offers promising certain income or requiring account credentials. If a syllabus is unclear, first ask for clarification; that differs from an explicitly false promise.

A trading loss does not oblige you to buy a course, and reading a book does not justify more risk. Material can wait until the relevant question arises. You can also finish studying a subject without applying it in the market.

A completed critical-reading card

Material: the Tharp edition identified above, covering R-multiples and position size. Reader's question: does “I am risking PLN 100” mean I cannot lose more? The publisher's catalogue confirms the subject coverage; the calculation below is an original synthetic example, not a quotation or data from the book.

On a small screen, the table scrolls horizontally. Focus it with the keyboard and use the arrow keys.

Synthetic example — R₀ recorded before each trade
TradePlanned R₀Net resultResult / R₀
APLN 100−PLN 150−1.5R
BPLN 100+PLN 200+2R
CPLN 100−PLN 50−0.5R
  • Working definition: R₀ is the positive initial amount of planned risk, and the numerator is the net result after costs. I do not change the denominator after seeing the result.
  • Observation: trade A lost 1.5R, more than the planned PLN 100. Total results are PLN 0, and the mean of the three R outcomes is 0R.
  • Limit of the conclusion: three invented observations check the arithmetic, not the strategy's expectancy. We do not know the frequency of future losses or the distribution of results.
  • What needs checking: the cost definition, when R₀ was recorded and order-execution conditions. If R₀ is missing or non-positive, I do not calculate the ratio; I record that a valid basis is missing.
  • Decision: I can add unambiguous journal fields and check their consistency using synthetic data. There is no basis for changing the strategy or position size, or for starting to trade.

The next resource can use the same card fields: source and edition, question, claim or definition, your own check, result, limitations and decision. If you cannot reproduce the method, record that gap. “I do not know” is more accurate than attributing an unverified conclusion to the author.

Reading card: planned R is not a loss cap. An original synthetic example. R₀ = PLN 100 recorded before each trade.
An original synthetic example. R₀ = PLN 100 recorded before each trade. With R₀ ≤ 0 or no record, do not calculate the ratio. Three invented observations do not support a strategy.
From reading to your own check
From reading to your own checkAfter reading, explain the mechanism in your own words, apply it to a new example and compare the result with the criterion. Record errors and limits rather than assuming mastery.1. Read the passageRecord the question, assumptionsand source.2. Recall without lookingExplain the mechanism inyour own words.3. Change the exampleUse different data ortest a boundary case.4. Compare and correctCheck the criterion; recordthe error and limit.

A proposed study method, not proof that a strategy works. A completed exercise concerns its checked scope, not all professional competence.

One resource and the next step

A short cycle for working with material
  1. Name the gap and choose the relevant chapter.
  2. Check the edition, legal access and prerequisites.
  3. Complete a small task outside a live account.
  4. Separate the exercise result from a claim about the market.
  5. Decide: return to the basics, investigate further, document understanding without changing anything, or close the subject.

There is no single required number of hours, books or trades after which knowledge becomes sufficient. Scope depends on the task and evidence of its completion. In the post-course development plan also allow for rest, changing your interests or continuing education without trading. This closes the course; it is not a commitment to another stage of risk.

Frequently asked questions

Must I read all the listed books?
No. Choose a resource or chapter addressing your current knowledge gap. The list is neither a mandatory canon nor a condition for starting or continuing to trade.
Is an older book still useful?
It may explain enduring concepts, but market descriptions, data and rules may be outdated. Check which parts need current documentation. The publication date alone does not determine quality.
Where can I legally find free materials?
The authors' site for An Introduction to Statistical Learning provides a PDF and materials, while GFXC publishes the FX Global Code. For other books, use the publisher, a bookshop or a library. A catalogue link does not mean full-text access.
Should an exercise after reading use a live account?
No. It can be a calculation, analysis of a synthetic example or reproduction of a lab. Such a result does not automatically justify implementing a strategy or increasing risk.
Does R limit a loss to 1R?
No. Initial planned risk is a reference point. The realised loss can be larger. An R outcome requires a correctly recorded positive denominator and unambiguous cost treatment.
What if reading gives no basis for changing what I do?
Record the concept understood, the limit of the conclusion and any further question. Keeping a strategy unchanged or ending further research are valid learning outcomes.

Bibliography and legal access

Catalogues, author websites and institutions checked: 8 September 2026. The bibliography confirms the stated editions and publicly described scope; it does not claim a full review of every book. The tasks and card are original educational examples for this chapter.

  1. McGraw-Hill, Van K. Tharp, Trade Your Way to Financial Freedom, 2nd edition — edition details and table of contents.
  2. Oxford University Press, Larry Harris, Trading and Exchanges: Market Microstructure for Practitioners — catalogue, the year 2002 and chapter coverage.
  3. Global Foreign Exchange Committee, FX Global Code — official access to the December 2024 version and translations.
  4. Wiley-VCH, David Aronson, Evidence-Based Technical Analysis — first edition, 2006, ISBN and scope description.
  5. James, Witten, Hastie, Tibshirani, Taylor, An Introduction to Statistical Learning: with Applications in Python — Springer; authors' website with a legal PDF, courses and recordings and Python edition errata.

Your learning progress

Finished this lesson? Mark it yourself. Opening or scrolling a page never marks it complete.

Progress requires JavaScript and browser storage.

Saved only in this browser. No account or cross-device sync. Clearing site data can erase your progress.

Jarosław Wasiński

Editor-in-chief, ForexMechanics.com • Macro & FX analyst

Independent analyst and practitioner with over 20 years of experience in financial markets. Actively involved in Forex since 2007, with a focus on fundamental analysis, OTC market structure, and disciplined capital risk management.

  • Creator and editor-in-chief of MyBank.pl, providing reliable financial education since 2004.
  • Author of hundreds of in-depth market commentaries, structural analyses, and educational materials for traders and investors.
  • Advocate for market transparency, promoting education grounded in hard data and institutional reports (BIS, ESMA, CFTC).

All content is educational and informational only – it does not constitute investment advice or a recommendation. Remember: leveraged trading (Forex/CFD) carries a high risk of rapid capital loss.