Quant / Trading Reading List

Personal study plan for building better VibeBullish product. Ordered beginner → advanced. Started 2026-08-26. Keep "Reading notes" at the bottom current — one line per chapter: "Chapter says X; in VibeBullish that's Y / we don't do that."

Why these, not "trading" books

VibeBullish's open problems are methodological, not tactical:

Skip: Market Wizards / Livermore / chart-pattern / day-trading / options books. Fun, not evidence, different game.

Phase 1 — Basics (≈4–6 weeks, in order)

• Malkiel — A Random Walk Down Wall Street — why most signals are noise, what efficient markets actually claims, why beating SPY (the §19 +10pp gate) is a serious bar. Easy read.
• Vocabulary weekend (Investopedia / any primer) — make these second nature: alpha vs beta, Sharpe/Sortino/max drawdown, market cap, P/E, momentum vs mean reversion, factor, benchmark, survivorship bias, look-ahead bias.
• Georgia Tech CS7646 "Machine Learning for Trading" (free, Udacity/edX) — built for programmers without finance. Middle third = correct backtesting, orders, portfolio value, out-of-sample eval — an annotated cmd/backtest. Best single item for this profile.
• Ernie Chan — Quantitative Trading (the thin first one) — for solo software people; backtest pitfalls, transaction costs, simple strategies. Skim Excel/MATLAB parts.
• Andreas Clenow — Stocks on the Move — rules-based momentum ranking at 20–90d horizons; very close to pure_quant + ext_guard.

Phase 2 — Intermediate (after Phase 1)

• Gray & Vogel — Quantitative Momentum (then Quantitative Value) — evidence-based factor construction; why the most recent month is skipped (ext-guard = rediscovered short-term reversal); ranking + rebalancing at 20d/60d-ish horizons.
• Ernie Chan — Algorithmic Trading — mean reversion vs momentum, Kelly sizing, costs.
• Ilmanen — Expected Returns — where returns come from; which strategy categories deserve a book at all.
• Quantopian lecture archive (GitHub) / QuantConnect bootcamp — alphalens-style evaluation (IC decay, quantile returns) worth porting into the backtester.

Phase 3 — Advanced (3–6 months out)

• López de Prado — Advances in Financial Machine Learning — triple-barrier labeling, meta-labeling, purged/embargoed CV, Deflated Sharpe Ratio / probability of backtest overfitting (ch. 14). Most relevant advanced book: §19 "Sharpe > 2 on one book" with ~12 books running → best one looks good by chance.
  • Papers: Bailey & López de Prado, The Deflated Sharpe Ratio; Harvey, Liu & Zhu, …and the Cross-Section of Expected Returns (argues t > 3 given the factor zoo).
• Grinold & Kahn — Active Portfolio Management (first third) — IR = IC × √breadth; the formal version of "breadth protective, concentration hurts"; how much a 0.05-IC signal can ever earn.
• Harris — Trading and Exchanges — microstructure; why fills after a 15% run-up are worse. Skim.
• Aronson — Evidence-Based Technical Analysis — data-mining bias for someone building signal after signal.
• WorldQuant University MScFE — free, rigorous, long; only if wanting the credential.

Product side (parallel track)

Models already produce a real ranking signal; retail users don't buy Sharpe ratios.

• The Mom Test (Fitzpatrick) or Continuous Discovery Habits (Torres).

Concrete follow-ups the reading should unlock

• Deflated-Sharpe / number-of-trials adjustment to the §19 verdict in autopilot_graduation.
• Alphalens-style IC / quantile-return report in cmd/backtest.

Reading notes

• 2026-09-01 (from the market-level-gate discussion, pre-reading): Grinold & Kahn's fundamental law (IR = IC × √breadth) is the formal reason a market-timing gate is suspect for us — the ranker makes ~30 independent bets/month, a market gate is ONE bet and needs enormous accuracy to add what breadth adds for free. Also the reason the concentration studies kept failing. Read the first third when Phase 3 arrives.
• 2026-09-01: Moreira & Muir, "Volatility-Managed Portfolios" — vol clusters (forecastable), returns don't; so scale exposure to realized vol instead of gating entries on trend. In VibeBullish: the SPY 20d realized-vol tile already computes the input; the queued market-pause experiment should be reframed this way.
• 2026-09-26: Jump Trading, ICML 2026 expo talk "Multi-Agent System Design and Evaluation for Quantitative Finance" — talk says: point-in-time correctness is architecture (the backtest kink at the LLM's knowledge cutoff), a fancier harness must earn its complexity after cost/latency/review are counted, score traces not just the headline metric (their harness win only showed up in trace analysis), use a multi-metric scorecard with hard constraints, and multi-agent overfitting risk scales with tokens exchanged, not agent count. In VibeBullish that's the available_at PIT gap, the registry + backtester-first rule, the §19 gate + diagnostic layer, and a qualitative case (not a proof; Astra review 2026-09-26) for narrow, versioned LLM-role output schemas. Full notes (repo only, not on the site): research/talk-jump-icml2026-multiagent-quant-2026-09-26.md.

Video track (preferred format — books above are the fallback/reference)

Same phases, watchable instead of readable. ✅ = availability verified (2026-08-31, plus the Plain Bagel sub-items 2026-09-09). An entry with no ✅ is a pointer at a channel/topic, not a checked-to-exist title — treat it as a lead to browse, not a queue.

Phase 1 — Basics

• The Plain Bagel — optional on-ramp; skip if short on time. Finance vocabulary in 10-min bites by a CFA (Richard Coffin). ⚠️ Corrected 2026-09-09: the earlier "watch the alpha/beta, factors, short selling, market efficiency episodes" line named topics, not verified episodes — no such titled set was confirmed to exist. The channel's recent uploads are news commentary (trade deals, SpaceX IPO, prediction markets), not vocabulary; the evergreen explainers live in the older The Plain Bagel Basics playlist — browse that, don't search the channel front page. Two individually confirmed items:
  • ✅ The Grossman-Stiglitz Paradox (collab with Ben Felix) — markets can't be perfectly efficient, because price discovery has to be paid for. This is the theoretical license for what the paper books are attempting; pairs with the §19 "beat the p95 of random portfolios" bar.
  • ✅ A Short Explanation of Short Selling (TED-Ed, feat. The Plain Bagel) — relevant to the short-book decision already made (rejected, see improvement roadmap).
  • Everything else this entry used to promise (alpha/beta, factors, market efficiency) is covered at higher evidence density by Ben Felix below. If you only do one Phase 1 channel, do that one.
• Ben Felix — Common Sense Investing (YouTube) — evidence-based, paper-driven. Watch: The (Expected) Cost of Pyramiding into Winners-adjacent episodes on momentum, factor investing, "Do Stock Picks Beat the Market?". This is Malkiel-in-video-form; skip the book if you finish ~15 of these.
  • Specifically queued 2026-09-01 (from the market-gate discussion): his market timing episode(s) ("Market Timing: Why You Shouldn't…" family) — the evidence against binary in/out gates; and volatility-managed portfolios / volatility timing — Moreira & Muir's result that vol is forecastable while returns aren't, i.e. why the registry's "hold tight" idea should be vol-SCALING, not an SMA on/off switch.
• ✅ Yale ECON 252 — Shiller, Financial Markets (Open Yale Courses, free on YouTube + Coursera) — 23 lectures; Nobel laureate; how markets/insurance/securities actually work. Watch at 1.5×; skip the banking-history lectures if pressed.
• ✅ Georgia Tech CS7646 — Machine Learning for Trading (Udacity ud501, free; YouTube playlist) — still the #1 item: it was always a video course. Programmer-first; the backtesting/portfolio-value lessons are an annotated cmd/backtest. (2026 reviews note the videos are old — the methodology content is what matters and hasn't aged.)

Phase 2 — Intermediate

• Patrick Boyle (YouTube) — ex-hedge-fund PM, dry humor; his Applied Portfolio Management / Statistics for Quants playlists are a lightweight Grinold & Kahn substitute; also good "how funds actually blow up" case studies.
• QuantPy (YouTube) — Python implementations of backtests, portfolio optimization, risk metrics; closest video analog to Ernie Chan's books.
• Ernie Chan talks/webinars (YouTube, search "Ernie Chan quantitative trading talk") — 1-hr condensations of both his books.
• Dimitri Bianco — Fancy Quant (YouTube) — what quant work actually is, model risk, career-level intuition; good reality checks.

Phase 3 — Advanced

• López de Prado — "The 7 Reasons Most Machine Learning Funds Fail" (YouTube, ~1hr, multiple recordings e.g. QuantCon) — the AFML book compressed into one talk: backtest overfitting, deflated Sharpe, purged CV. Watch this even before Phase 3 — it's the §19 multiple-testing problem, from the source.
• MIT OCW 18.S096 — Topics in Mathematics with Applications in Finance (free video lectures, MIT OpenCourseWare) — the math-heavy tier: stochastic processes, portfolio theory, factor modeling. Only if the appetite develops.
• QuantConnect bootcamp videos / Quantopian lecture recordings (YouTube) — IC decay, quantile returns, alphalens-style evaluation to port into the backtester.
• ✅ Jump Trading — Multi-Agent System Design and Evaluation for Quantitative Finance (SlidesLive, public, 51 min; ICML 2026 expo page) — how a prop shop benchmarks LLM agents on market-analysis tasks: point-in-time eval cases mined from real analyst chats, baseline-first harness comparisons, trace-level scoring, and the "tokens exchanged bounds overfitting" argument. Done 2026-09-26 via transcript + slides; notes in research/talk-jump-icml2026-multiagent-quant-2026-09-26.md.

Watching habits that make it stick

• Same rule as reading: after each video that lands, one line under Reading notes above.
• Prefer 1.5–1.75× with pauses over 1× passive; passive watching retains ~nothing.
• When a video names a concept you've already discovered empirically (short-term reversal ≈ ext-guard, breadth ≈ concentration studies), that's the note to write down.