Book: Hedge Fund Alpha: A Framework for Generating and Understanding Investment Performance
Author: Multiple contributors; edited by John M. Longo
Difficulty: Advanced
The Big Idea
An attractive research idea is only the beginning. A fund must turn it into executable positions, finance those positions, measure the risks it actually takes and remain operationally sound long enough for the thesis to matter. This workbook connects those stages and asks where an apparent edge could disappear.
The book is a collection of specialist chapters, edited by John M. Longo and published by World Scientific in 2009, ISBN 9789812834652. Read the assigned chapters alongside this original companion. The book supplies its arguments, historical illustrations and research references; this workbook adds different calculations, evidence tests, decision exercises and explained answers. It does not recreate the original charts, tear sheets or sample background report.
Page references below mean the printed page numbers, not the PDF viewer count. Some pages near the appendices have different viewer offsets, so use the printed numbers or search the chapter title. Bibliographical details are taken from the supplied title and copyright pages. Named firms, platforms, compensation, regulations and market structures in the book describe its publication period.
If you remember one thing: alpha is a model-dependent assessment of performance, not a synonym for profit, leverage, a successful forecast or a sophisticated instrument.
What the source provides
Research arguments, country case studies, derivative trade sketches, execution methods, behavioral explanations and historical hypothetical portfolios. They vary in evidential strength and do not constitute a verified live fund record.
What this companion adds
Original examples and experiments, explicit assumptions and corrections where a formula or description is misleading. No country chapter is a current market-access guide, legal checklist or recommendation to trade there.
Quick revision sheet
| Concept | Formula or decision rule | Avoid this shortcut |
|---|---|---|
| Model-dependent alpha | Actual return minus the return expected under a stated risk model. CAPM illustration: Rp − [Rf + β(Rm−Rf)]. | A benchmark excess return is not automatically risk-adjusted alpha. Chapter 1. |
| Research mosaic | A documented inference assembled from appropriate evidence. | More information is not necessarily independent or lawful-to-use information. Chapter 2. |
| VWAP / TWAP | VWAP weights prices by traded volume; TWAP weights by elapsed time. | A volume-following schedule is not plain equal-time slicing. Chapter 8. |
| Expectancy | p × average gain − (1−p) × average loss magnitude − costs. | A high win rate can coexist with losses. Keep trade and daily observations distinct. Chapter 12. |
| Binary Kelly illustration | With a loss of the stake and a net win of b times the stake: f* = p − (1−p)/b. | This is a theoretical fraction at risk under restrictive assumptions, not a universal fraction of capital to invest. Correction. |
| Drawdown | At each time, compare value to its preceding running peak. Maximum drawdown is the largest such decline. | Do not pair a trough with a peak that occurred later. Risk measurement. |
| Exposure | Simple equity gross = (long + absolute short)/NAV; net = (long − absolute short)/NAV. | Zero net dollars need not mean zero beta, liquidity risk or option risk. Exposure drill. |
| VaR / expected shortfall | VaR is a loss quantile for a stated model, horizon and confidence. Expected shortfall describes the severe-loss tail. | VaR is not a worst-case loss cap, and not every VaR method assumes normality. Tail drill. |
| Liquidity ladder | Compare dated cash obligations with realistically accessible assets under stress. | A notice date or request frequency does not guarantee cash on the payment date. Fund-of-funds lesson. |
| Lifecycle | Separate hypothesis failure, implementation deterioration and business strain. | Recent underperformance alone does not identify which one occurred. Chapter 14. |
Seven study blocks and chapter map
Allow more than one sitting for the heavier blocks. Attempt answers before expanding the keys. Each completed block should leave a piece of evidence in your notebook; finishing the plan is a study milestone, not a promise of profitable trading.
| Block | Assigned source reading | Your deliverable |
|---|---|---|
| 1 | Preface; Ch. 1 Introduction, pp. 3–16; Ch. 2 Hedge Fund Research vs. Traditional Research, pp. 17–32. | An alpha-model statement and timestamped research mosaic. |
| 2 | Ch. 3 Brazil, pp. 33–50; Ch. 4 Russia, pp. 51–66; Ch. 5 India, pp. 67–84; Ch. 6 China, pp. 85–102. | Four separate country hypothesis cards plus an implementation comparison. |
| 3 | Ch. 7 Using Derivatives to Create Alpha, pp. 103–118; Ch. 8 Best Execution of Hedge Fund Strategies, pp. 119–136. | A derivative risk decomposition and execution-cost ledger. |
| 4 | Ch. 9 Growth of the Hedge Fund Management Company, pp. 137–152; Ch. 10 Fund of Hedge Funds, pp. 153–176. | Business-growth decision, layered fee bridge and liquidity ladder. |
| 5 | Ch. 11 The Psychology of Hedge Fund Managers, pp. 179–192; Ch. 12 Risk Management for Hedge Funds, pp. 193–220. | A behavioral pre-mortem, corrected sizing calculation and risk dashboard. |
| 6 | Ch. 13 Hedge Fund Due Diligence, pp. 221–248; Ch. 14 The Lifecycle of a Hedge Fund Investment Strategy, pp. 249–264. | Evidence-based diligence decision and a predeclared strategy-review protocol. |
| 7 | Ch. 15 Seven Emerging Trends, pp. 265–280; Appendix 1, pp. 283–292; Appendix 2, pp. 293–297; Appendix 3, pp. 299–311. | Forecast audit, resource plan, background-evidence exercise and capstone. |
The prefatory biographies explain contributor perspectives; they are not independently verified current credentials. The index is a navigation tool. The map includes the complete substantive text and all three appendices; blank separator pages do not require exercises.
Research and the meaning of alpha
1. Choose the explanation before celebrating the return
Read: Chapter 1, pp. 3–16. Goal: Distinguish risk exposure, a performance residual and evidence of skill.
The opening chapter offers several possible paths to alpha: a better research synthesis, information and execution. Its short strategy descriptions identify different return mechanisms rather than one uniform “hedge fund” product. Even two funds with the same label may have different exposures.
An expected-return model matters. A residual can change when the benchmark or factors change. A positive residual in one period is not proof of repeatable skill; it may reflect chance, an omitted factor, option exposure, stale valuations or a mismatch in observation dates.
Use the source’s strategy list as a reading assignment: for each category, identify what is bought, what is sold, the proposed source of return and the event that could cause loss. An arbitrage label does not remove settlement, financing, execution or model risk.
Original worked example
Our fictional fund returns 12%, while the risk-free reference is 3% and the market returns 10%. With beta 0.8, the single-period CAPM expected return is 3% + 0.8 × 7% = 8.6%; the residual is 3.4 percentage points. If beta is instead 1.2, expected return becomes 11.4% and the residual is only 0.6 points. Neither number establishes statistical skill.
Your turn: Why would simply subtracting the market’s 10% produce a different answer? For a merger trade, a macro trade and a short-biased fund, name one distinct failure mechanism.
Open the explained answer
The 2-point market-relative result assumes a comparison that does not adjust for the stated beta and risk-free reference. Merger trades face deal failure; macro trades face a wrong or mistimed economic view; short-biased funds face price rises, borrow costs and recalls. Additional factors may still be needed in every case.
Specify the model and the unit. Percentage-point residuals are estimates, not certificates of skill.
2. Build a mosaic that can survive disconfirming evidence
Read: Chapter 2, pp. 17–32. Goal: Turn a research dashboard into a documented inference.
The chapter contrasts time horizons, short-side research, flexible mandates and multi-instrument analysis. Its product, channel-check, search and expert-network examples show ways to form an investment view. They do not prove that every clue predicts returns, and several clues may originate from the same underlying report.
Separate component-cost estimates from a complete accounting gross-margin estimate: distribution, warranty, recognition and other costs can matter. Record publication time and availability, not just the fiscal period described. A study using later-released information at an earlier decision date introduces look-ahead.
The source’s discussion of FOIA and information access needs care. A refusal or neither-confirm-nor-deny response does not establish misconduct. The SEC’s FOIA guide describes exemptions and such responses. CFA Institute’s information standards distinguish appropriate mosaics from acting on material nonpublic information. A research label or paid expert conversation does not make restricted information usable.
Original worked example
A fictional manufacturer reports 20% revenue growth, distributors report rising unsold stock, and receivables rise 40%. One explanation is aggressive revenue recognition; another is a genuine expansion with temporary working-capital needs. Write both hypotheses, then request cash collection, returns, contract terms and historical seasonality before choosing one.
Your turn: Design a five-field dashboard for that case. What would make three apparently independent web stories count as one piece of evidence?
Open the explained answer
Use hypothesis, observation/source, public-availability timestamp, alternative explanation and next falsifying test. If all three stories quote one distributor interview or one press release, their common origin means they are not three independent confirmations. Preserve provenance and uncertainty rather than scoring repeated headlines as stronger evidence.
A useful dashboard exposes uncertainty. It should help change your mind, not decorate your initial belief.
Country case-study laboratory
Chapters 3–6 connect market institutions, investor behavior and strategy hypotheses. Their growth forecasts, exchange names, shorting rules, share-access regimes and instrument descriptions are dated. Study the method of making a claim conditional on infrastructure. Do not treat historical descriptions of national culture as fixed traits of every investor, or as proof of an exploitable effect.
3. Brazil: high nominal yields need a base-currency test
Read: Chapter 3, pp. 33–50. Goal: Connect inflation, currency, liquidity and a strategy hypothesis.
The Brazil chapter emphasizes the inflation legacy, regional economy, market infrastructure and investors’ liquidity preferences. Its proposed macro, credit, event, statistical and multistrategy approaches are forecasts by the contributors. Volatility creates potential losses as well as potential opportunities.
Treat quoted GDP growth carefully: nominal versus real, local-currency versus dollar measures and purchasing-power adjustments are not interchangeable. Likewise, a yield is neither a total-return forecast nor a base-currency return. Modern access, trading rules and fund liquidity terms require separate current verification.
Original worked example
A fictional local-currency bond investment gains 14% including distributions, while that currency loses 12% against the investor’s base currency. The unhedged base-currency return is 1.14 × 0.88 − 1 = 0.32%, before fees and taxes. Local inflation of 8% would imply a local real return of 1.14/1.08 − 1 = 5.56%; that is a different question.
Your turn: Choose one proposed strategy from the chapter. Write a falsifiable hypothesis and name two infrastructure requirements. What if assets take ten days to sell but investors expect rapid withdrawals?
Open the explained answer
An event hypothesis could specify the sign and horizon of a post-announcement return relative to a matched benchmark. It requires timestamped news, suitable price data and executable access including borrow for short trades. A ten-day sale horizon against faster withdrawals creates a liquidity mismatch; a plan must address cash, contractual terms and stressed exits. Historical statements about daily liquidity are not current universal rules.
Match the return to the investor’s currency and cash needs. A high headline yield can disappear in translation.
4. Russia: distinguish a price discrepancy from an executable trade
Read: Chapter 4, pp. 51–66. Goal: Identify political, settlement and market-access dependencies.
The historical chapter links commodities, banking development, exchange fragmentation and political risk to its proposed strategies. The transferable lesson is that both legs of a price comparison need to be tradable, deliverable and financeable. A halted or inaccessible leg can turn apparent convergence into an open risk.
Do not carry the chapter’s 2008 market-access and legal assumptions into a live trade. For example, current Russia-related sanctions must be assessed separately; the OFAC program pages are one relevant U.S. official source. Applicable jurisdictions, counterparties and licenses matter. This exercise does not determine the legality or availability of any transaction.
The chapter’s behavioral studies are context-specific research claims. Their samples and dates need evaluation; they are not a basis for predicting an individual’s behavior from nationality.
Original worked example
A fictional asset is quoted at 100 on one venue and 104 elsewhere in the same currency and unit. If estimated round-trip charges are 1.5 and financing/transfer costs are 1.0, the apparent 4-unit difference leaves 1.5 before uncertainty. If the second venue’s price is stale or shares cannot be transferred or shorted, even that calculation is not an executable arbitrage.
Your turn: List four checks required before treating the difference as realizable. How does a trading halt alter a nominally hedged position?
Open the explained answer
Check instrument equivalence and conversion ratio, synchronized executable quotes and size, legal access and borrow, and settlement/transfer/financing terms. A halt can prevent one leg from being opened, adjusted or closed while the other moves. The residual exposure, funding and settlement risks remain; a chart cannot certify that both legs could have traded.
Execution permission and settlement are part of the payoff. Prices alone are insufficient.
5. India: an index hedge leaves an active bet
Read: Chapter 5, pp. 67–84. Goal: Evaluate imperfect hedges and ownership-event research.
The chapter discusses ownership concentration, investor access, P-notes, depositary receipts, hedging constraints and several strategy families. Its descriptions of foreign-investor categories and convertible-bond terms are historical generalizations, not instructions for any current security. Read actual contract and offering terms.
A family-controlled issuer or rights offering can motivate a research question without proving a profitable rule. Statistical significance, transaction timing, sample selection, shortability and costs must be checked. A subgroup result is not a guaranteed outcome for a single company.
An index future may reduce broad equity exposure while leaving sector, issuer, currency and liquidity exposures. The selected index might not represent the long book well.
Original worked example
Our fictional portfolio has $2 million of local equity exposure with estimated index beta 1.3. A simplified beta hedge uses $2.6 million of short index exposure. If the index falls 5%, the linear model predicts −$130,000 on the book and +$130,000 on the hedge. An additional issuer-specific −4% shock costs $80,000 on the long book and is not offset by that model hedge.
Your turn: Why is $2 million of short index exposure a different hedge? Specify a defensible test of the rights-offering hypothesis without using hindsight.
Open the explained answer
An equal-dollar hedge leaves estimated $0.6 million of beta-adjusted index exposure under the model. For the rights test, predefine ownership classification using then-available filings, announcement time, observation windows, benchmark, corporate-action adjustments, borrow and costs. Include all eligible events and failed/delisted firms. A valid study may reject the hypothesis; do not require a favorable result.
Hedge the specified risk and name what remains. Equal dollars do not establish equal sensitivity.
6. China: a segmented price is not necessarily a free profit
Read: Chapter 6, pp. 85–102. Goal: Separate access constraints, accounting quality and behavioral hypotheses.
The China chapter compares different trading venues and share classes, accounting effects and investor reactions to news. Its multiple-market discussion should not be read as one uniform legal or investment-access regime. A premium between related shares may persist when conversion, access, capital movement or shorting differs.
Earnings that depend on gains in other companies’ shares can transmit market swings into reported profit. Reconcile operating and non-operating contributions rather than treating all earnings growth as improved product economics.
The suggestion that a prominent state-linked company will receive support is a hypothesis, not a guarantee for its equity or debt holders. Support for operations can coexist with dilution or creditor losses. Nor can a negative-news reversal be implemented by buying at “the bottom,” a point identifiable only afterward.
Original worked example
A fictional company earns 60 million from operations and 40 million from investment gains: total 100 million. Next period operations rise to 66 million but investments lose 20 million. Total profit falls to 46 million, a 54% decline, despite 10% operating growth. A headline earnings model that ignores the investment book can miss the driver.
Your turn: Write an implementable reversal definition for study and one reason it is only a proxy for the chapter’s behavioral idea. What evidence would a government-support thesis need?
Open the explained answer
For example, study returns beginning after a fixed five completed trading days following a timestamped negative announcement; this avoids choosing the eventual trough. It remains our definition, not evidence of overreaction or investor motive. A support thesis needs specific legal, capital-structure and policy evidence; even support does not identify which security benefits. Classification by national culture is not a substitute for a test.
Name the mechanism and the security that benefits. A broad growth or support story does not specify an investor payoff.
Derivatives and execution
7. Decompose the derivative before calling the result alpha
Read: Chapter 7, pp. 103–118. Goal: Identify directional, volatility, credit and funding exposures.
The chapter sketches several ways to express a view or connect markets. A derivative can reduce one exposure while creating another. A profitable directional rate trade is not automatically “pure alpha,” and a volatility skew does not establish mispricing merely because it differs from a constant-volatility model.
Delta describes local sensitivity, not a permanent hedge ratio. Gamma, volatility, rates, credit and liquidity can change the result. Exchange-traded prices help price discovery but do not remove model risk in valuation interpretation, hedging and risk measurement.
Keep portable alpha conditional: a manager position plus a hedge against its original market and a new market exposure may approximate a desired combination, but beta estimation, basis, collateral and financing matter. The alpha portion must first exist after costs.
Original worked example
Our fictional option position has a local dollar delta of $60,000. A 1% underlying rise implies approximately +$600 from the first-order delta term. If gamma and volatility also change, the actual option change differs. Paying a $4,000 premium does not make $4,000 the notional directional exposure, and a $60,000 delta is not a worst-case loss estimate.
Your turn: An equity manager has estimated $1.1 million market-equivalent exposure. What would a $1 million index short leave in the simple model? Why might the “portable” result still fail?
Open the explained answer
It leaves $100,000 of estimated market-equivalent exposure. Beta changes, index mismatch, residual factor exposures, hedge costs and margin demands can overwhelm the desired residual. A realized residual is not proof of persistence, and imperfect local hedges need stress tests beyond small price moves.
Decompose the risk before judging the edge. A different wrapper does not manufacture skill.
Eleven derivative sketches: what must be verified?
After reading Chapter 7’s trade sketches, complete this original challenge: for each row, name the missing assumption before reading the final column. The table is a risk-analysis drill, not an instruction to place these trades.
| Reading topic | Original challenge | What a sound answer must address |
|---|---|---|
| Futures across venues | Identical-looking contracts differ by 0.3%. Can the gap be captured? | Contract units, delivery, dates, currency, executable size, simultaneous access, margin and costs. Similar labels need not mean identical delivery obligations. |
| Realized versus implied volatility | Forecast volatility is 25%; an option implies 20%. Is the difference a guaranteed profit? | Forecast error, option premium, delta-hedging path and costs, jumps and volatility exposure. It is a hypothesis, not a locked spread. |
| Volatility skew | A low-strike option implies higher volatility. Is the model wrong or the option cheap/expensive? | Market pricing of tails and non-lognormal outcomes may explain skew. A skew trade can retain jump and nonlinear exposure. |
| Cash-futures basis | The futures price exceeds cash. Is that sufficient? | Financing, distributions or income, delivery options, contract multipliers, settlement and carry-adjusted fair value. |
| Optionality in bonds | A callable bond offers a higher coupon. What was sold to the issuer? | The issuer’s exercise right can limit upside and alter duration. Hedge instruments may have different exercise dates or underlying terms. |
| Convertible bonds | A long convertible is paired with a stock short. What survives? | Credit/default, rates, volatility, changing delta, borrow and liquidity risks. A long put or short call alone is not identical to a short stock position. |
| Macro rates | A rate forecast proves correct, but the position loses. How? | Timing, curve segment, carry, instrument choice, priced-in expectations and costs. Correct forecasting and model-adjusted alpha differ. |
| MBS versus Treasuries | Initial durations match. What can move them apart? | Prepayments, changing convexity, spread and model error; a duration hedge is local and incomplete. |
| Credit derivatives | An index hedges a specific credit pool. What could remain? | Collateral quality, recovery, default dependence, tranche structure, counterparty and index-basis mismatch. |
| Energy optionality | A delivery contract allows quantity changes. What is missing from a plain price hedge? | Volume flexibility, seasonality, spikes, physical capacity and correlated operating risks. |
| Long/short allocation | Half the allocation in a daily 2x product is described as equivalent exposure. Over what horizon? | At best a stated daily exposure approximation; resetting, financing, volatility, tracking and compounding can change multi-day outcomes. |
Correction: daily leverage is not a fixed multi-day multiple
Chapter 7’s leveraged-fund allocation example needs a horizon qualification. For a fictional index that rises 20% then falls 16⅔%, $100 returns to $100. A perfect daily 2x product, ignoring all costs, goes from $100 to $140 to about $93.33. Daily objectives were met, yet the multi-day result was −6.67%. See the SEC’s leveraged and inverse ETF bulletin. Also distinguish a gold-only exposure from a diversified commodity allocation; one asset does not represent the full commodity universe.
8. Separate execution benchmarks from execution promises
Read: Chapter 8, pp. 119–136. Goal: Measure implementation shortfall without double counting.
The chapter’s market-history discussion motivates a continuing process: plan the trade, monitor it and reconcile the result. Its venue names, regulations and historical market shares are not a current routing directory. An OMS or EMS can manage instructions and reports; submitting or canceling an order is not the same as confirming a fill or cancellation.
VWAP is a volume-weighted price over a specified window. Plain TWAP weights time; a simple equal-time schedule is not a volume-proportional schedule. The chapter’s TWAP paragraph mixes these ideas. A participation-of-volume approach follows volume instead, and does not guarantee completion or price. Broker-specific algorithms add their own conditions; IBKR’s TWAP documentation is one concrete example.
A dark venue may reduce information display, but anonymity, no leakage and low impact are not guaranteed. Nonexecution and adverse selection still matter. A limit order controls a price condition, not the certainty of receiving the desired quantity.
Original worked example
A fictional buy decision is made for 10,000 shares at $50. Only 6,000 fill at $50.10; the remaining 4,000 are still unfilled when the evaluation price is $50.40. With $30 explicit fees, shortfall relative to the decision-price paper position is 6,000×$0.10 + 4,000×$0.40 + $30 = $2,230. On the $500,000 intended notional, that is 44.6 basis points.
Your turn: Why does comparing only the filled average with market VWAP miss part of this result? Should you add a separate spread estimate after already measuring execution against the decision price?
Open the explained answer
The unfilled opportunity component is absent from a fills-only comparison, and the chosen VWAP window may differ from the decision time. Execution-price deterioration already incorporates spread and other price effects; do not blindly add estimated components again. Use a consistent decomposition with explicit fees and clearly stated opportunity valuation. This example does not identify how much was caused by the order itself.
Reconcile intended quantity, actual fills and the unfilled balance. A benchmark is a comparison, not a guarantee.
| Method | Our small example: prices $40, $42, $44; volumes 100, 300, 100 | Interpretation |
|---|---|---|
| Equal-time sampled TWAP | ($40+$42+$44)/3 = $42 | Each equally spaced observation has equal weight. Real continuous-time definitions may use finer observations. |
| Sample VWAP | (40×100+42×300+44×100)/500 = $42 | Here the same value is coincidental. Change the last volume to 600 and VWAP becomes $43.00, while sampled TWAP stays $42. |
| Participation target | 10% of 500 observed shares = 50 shares | A volume participation amount is not a return or a guaranteed executable fill. |
The investment business
9. Growth can diversify revenue while straining competence
Read: Chapter 9, pp. 137–152. Goal: Compare related expansion, unrelated expansion and external seeding.
The growth chapter asks how a firm can move beyond one opportunity set. Related strategies may reuse expertise but retain shared factor risks. An apparently uncorrelated strategy may diversify historical returns while introducing skills, infrastructure and governance the original team lacks.
A multistrategy vehicle and several separate funds create different oversight, fee and liquidity arrangements. The chapter’s investor-slot and look-through descriptions are not a current legal test. More vehicles do not create unlimited investment capacity, and low historical correlation does not guarantee that one sleeve will gain in every crisis.
Seeding also separates two assets: the investment in the fund and an interest in its management business. They have different cash flows, transfer rights and potential lockups. Historical ownership “rules of thumb” are not standard contract terms.
Original worked example
A fictional firm has $100 million in a process planned around buying no more than 5% of an asset’s ordinary daily volume. Doubling assets without changing weights doubles the order participation to 10% unless execution is spread over more time. A related new fund using the same names may compete for the same capacity rather than create new capacity.
Your turn: Create a two-option growth memo: add a related sleeve internally or seed an outside team. Name one capability test and one conflict for each.
Open the explained answer
For the internal sleeve, test staff, systems and aggregated capacity; review cross-fund trade allocation and expense sharing. For external seeding, test independent operational capacity and due-diligence access; review revenue-sharing, preferential rights, withdrawal and control terms. Evaluate current contracts and avoid equating an attractive business investment with an attractive fund allocation.
Grow the capability before the promise. A larger firm is not automatically a stronger investment process.
10. A fund of funds adds a service layer and a risk layer
Read: Chapter 10, pp. 153–176. Goal: Analyze fees, look-through exposures, liquidity and wrappers together.
The chapter covers manager sourcing, allocation, monitoring and portfolio construction, then compares direct holdings, a pooled fund of funds and separate accounts. Access and research can be valuable services. The service must be evaluated against its fees and the risks it actually reduces. Multiple managers can still share counterparties, crowded positions or illiquid exposures.
A managed account can improve visibility and control depending on its arrangements; it does not eliminate fraud, valuation error, trading loss or liquidity risk. A registered product is not an endorsement. The chapter’s tax rates and proposed structured-product tax outcomes are dated illustrations, not applicable tax conclusions.
A structured product can add issuer credit risk, caps, conditions and a maturity mismatch. “Principal protection” requires reading exactly who promises what, when and subject to which conditions. An overlay may alter price exposure without releasing cash from a locked underlying fund.
Original worked example
Our simplified convention: an underlying fund earns 12% gross on $100, charges $2 management fees and 20% of the remaining $10 profit ($2), ending at $108. A top fund-of-funds layer then charges $1 on opening capital and 10% of its remaining $7 profit ($0.70). Investor ending value is $106.30, before any other expenses or taxes. The convention is ours; actual documents may specify different bases and timing.
Your turn: A fictional fund of funds has $20 million NAV, $2 million cash and $5 million of investor payments due in 30 days. Its expected next underlying redemption proceeds are $4 million in 90 days. Calculate the 30-day gap and propose the next research action.
Open the explained answer
The 30-day gap is $3 million. Ninety-day proceeds cannot simply be counted as cash in 30 days, and could be subject to gates or suspension. Verify the actual obligations and liquidity terms, additional realizable assets, financing availability and stress assumptions. Do not assume borrowing is always available or solves an ongoing structural mismatch.
Look through the wrapper. A new fee or legal layer does not erase the underlying exposures.
Layered leverage exercise
For an original simplified financing example, a fund of funds uses $100 equity and $100 borrowing at 5% to hold $200 of underlying investments. If those investments lose 8%, equity loses $16 plus $5 interest, leaving $79: a 21% loss before other fees. Any leverage inside the underlying funds adds another source of exposure not captured by the 2x top-layer asset/equity ratio.
Your turn: Under the same one-period assumptions, what if underlying investments gain 8%? Does that establish that leverage adds alpha?
Open the explained answer
The gain is $16 minus $5 interest = $11, or 11% on opening equity. It amplifies a favorable outcome and changes risk. One favorable arithmetic outcome does not establish superior risk-adjusted performance. Margin, changing financing costs and path-dependent liquidation are omitted.
Behavior and risk
11. Study incentives without diagnosing motives from returns
Read: Chapter 11, pp. 179–192. Goal: Link observable behavior to controls and alternative explanations.
The chapter proposes how compensation, benchmarks and monthly reporting can affect behavior. Its emotional ranking is a conceptual account, not a measured personality test. A smooth return series may reflect illiquidity or valuation practices; it is a reason to investigate, not proof of fraud.
The behavioral concepts overlap. Confirmation and availability affect evidence selection; conservatism and disposition affect updating and exits; endowment, sunk cost and loss aversion affect attachment. Hubris, overconfidence and illusions of knowledge/control affect sizing. Recognition, recency/overreaction, small samples and self-attribution can distort evaluation. These labels should generate questions about decisions rather than confident claims about someone’s mind.
Coinvestment may partly align interests, but it does not cancel a convex incentive fee or ensure prudent control. A manager can have personal money at risk and still take excessive exposure.
Original worked example
A fictional manager receives 20% of positive profit and nothing below zero, with no prior high-water-mark adjustment in this isolated example. Outcomes of +$2 million or −$2 million are assumed equally likely. Investor gross expected profit is zero; expected incentive fee is 0.5×$400,000 = $200,000. This simplified asymmetry shows an incentive, not a forecast of behavior.
Your turn: Write a review question and a control for each: ignoring negative distributor evidence, doubling after losses and claiming all wins are skill. How would you test return smoothing?
Open the explained answer
Ask for contrary evidence and use a preassigned independent challenger; inspect exposure changes and enforce documented authority/limits; compare prior probabilistic forecasts with outcomes and record both luck and error. For smoothing, examine independently supported marks, revision history and higher-frequency economic exposures. Serial correlation or smooth returns alone cannot identify the cause.
Translate a bias into an observable decision and a control. Awareness alone does not fix the process.
12. A winning percentage is not an edge
Read: Chapter 12, pp. 193–202. Goal: Combine frequency, payoff, costs and sizing assumptions.
The chapter links P&L analysis to trading risk. Its Impact Ratio is average gain divided by average loss magnitude. The win frequency and this ratio must be evaluated together. No universal 60% winning threshold is required for profitability; a low win-rate approach can be viable with sufficiently large gains, and a high win-rate approach can hide severe losses.
Keep the observation unit consistent. Five positive days out of eight days is a daily batting average, not a count of winning trades. Several trades can be embedded in each day, and dependence between observations reduces the information in a headline count. An impact ratio alone also cannot establish whether exits were slow or why losses occurred.
Original worked example
Our fictional trade outcomes win 40% of the time, with average gain $300 and average loss $100. Gross expectancy is 0.4×300 − 0.6×100 = $60 per trade. With average costs of $15 it becomes $45. A second process wins 80% with $50 average gains and $300 average losses: 0.8×50−0.2×300 = −$20 before costs.
Your turn: For the first process, compute the gross break-even win rate. Why is averaging a small sample of gains and losses insufficient to recommend a size?
Open the explained answer
Break-even p satisfies 300p−100(1−p)=0, so p=25% before costs. A small sample may misestimate p, payoff distributions, dependence, gap losses and tail events. Costs and changing market conditions matter. An average loss is not a guaranteed maximum loss, so substituting it into a sizing formula can materially understate risk.
Count, magnitude and costs belong together. Then ask how reliable those estimates are.
Correction: the Kelly payoff ratio is inverted in the source
In the reading
Printed p. 200 multiplies loss probability by average gain divided by average loss, then subtracts that amount from win probability. Its stated 25% result follows that printed expression, but that expression is not the standard binary Kelly formula.
Correct theoretical definition
If a losing bet loses the entire stake and a winning bet gains b times the stake, expected log growth is p ln(1+bf) + (1−p) ln(1−f). Differentiating and setting it to zero gives f* = p − (1−p)/b, subject to feasibility and the model assumptions.
For our different fictional binary gamble, p = 0.6 and b = 2. The formula gives f* = 0.6 − 0.4/2 = 0.4. This means a theoretical 40% fraction at risk in that idealized gamble, not an investment recommendation. Assumed exact probabilities, fixed payoffs, repeated opportunities and no extra frictions are strong conditions. An ordinary security position usually does not lose exactly its invested amount with a known probability, and multiple correlated positions require a portfolio model.
Your turn: Keep b = 2 but reduce estimated p to 0.4. What changes, and why is that sensitivity relevant?
Open the explained answer
The theoretical fraction becomes 0.4−0.6/2 = 0.1, or 10%, under the same idealized model. A probability-estimation error changed the result by 30 percentage points. The calculation does not validate p, limit real losses or justify using full Kelly on uncertain trading estimates.
13. Risk statistics answer different questions
Read: Chapter 12, pp. 201–220. Goal: Use definitions that preserve chronology, units and assumptions.
Drawdown uses an earlier running peak. Subtracting the minimum and maximum of an entire unordered sample can join dates in the wrong order. Standard deviation of account dollar values is also different from volatility of returns: size and cash flows affect the former.
Correlation measures linear co-movement. Independence implies zero correlation when the relevant moments exist, but zero correlation does not generally imply independence. A constant return series has undefined correlation rather than a meaningful zero. Beta is benchmark sensitivity, not total volatility or a worst-case loss measure.
The source’s risk statistics should be treated as a toolkit, not an automatic ranking. Sharpe, Treynor, Jensen, Sortino, Omega and VaR depend on definitions, horizons and data. A government bond held before maturity can have price, inflation and currency risks; calling all Treasury bonds riskless is too broad.
Original worked example
For a fictional cash-flow-free value path 100 → 80 → 120 → 110, the largest chronological decline is 20%, from 100 to 80. Pairing the later 120 high with the earlier 80 low gives 33.33%, but that is not a realized peak-to-later-trough drawdown on this path. The final drawdown is 8.33% from 120 to 110.
Your turn: A portfolio is equal dollars long and short. Can its beta and option exposure be zero by assumption? What would you put beside any risk statistic in a review report?
Open the explained answer
No. Estimate position sensitivities and their assumptions; use a stress analysis for nonlinear and basis exposures. Report dataset and dates, frequency, currency, fee and cash-flow treatment, benchmark/target, formula convention and uncertainty. A single number cannot establish investor suitability or predict a guaranteed loss boundary.
Preserve time order and the denominator. A familiar formula name does not protect against a wrong calculation.
Exposure and sensitivity drill
| Question | Worked answer | Limit |
|---|---|---|
| $10m NAV; $15m long; $10m short | Gross = 250%; net = 50%. If long beta is 1.2 and short beta 0.8, net dollar beta is $18m−$8m = $10m, or 100% of NAV. | Single-factor linear approximation; betas can change. |
| Three long option contracts; multiplier 100; underlying $80; delta 0.5 | Dollar delta = 3×100×80×0.5 = $12,000. A 1% small move implies about $120 from delta alone. | Use signed positions and actual contract terms. Premium and notional differ; gamma, vega and gaps remain. |
| 200,000 shares; ordinary volume 400,000/day; assumed participation 10% | Planning liquidation time = 200,000/(400,000×0.1) = 5 trading days. At half volume, it becomes 10. | Constant volume and participation assumptions do not guarantee completion or negligible market impact. |
| Sector, country and product buckets | Classify each position consistently, then aggregate common economic drivers and counterparties. | Sector labels, credit ratings and seniority do not by themselves measure all loss risks. |
Match each performance measure to its assumptions
| Measure | Our explicit working convention | Interpretation boundary |
|---|---|---|
| Sharpe | Mean periodic excess return / standard deviation of periodic excess return. | Match the risk-free reference and frequency; nonnormality and stale marks matter. |
| Treynor | Mean return above the risk-free reference / estimated benchmark beta. | A beta-based ratio omits idiosyncratic risk. Near-zero or negative beta makes naïve rankings problematic; it is not a general risk measure for an undiversified account. |
| Jensen alpha | Intercept in a specified excess-return market regression, or a clearly labeled single-period CAPM residual. | A model residual, dependent on sample and specification; not evidence of persistent skill by itself. |
| Sortino | Mean return above the chosen target / target downside deviation; here downside deviation = sqrt(mean(min(R−target,0)2)) over all observations. | Conventions differ. State target and denominator; standard deviation of only losing observations is a different calculation. |
| Omega | For equally weighted observations at target T: sum(max(R−T,0)) / sum(max(T−R,0)). | Target-dependent; undefined/infinite when the loss denominator is zero. A short sample may simply miss the tail. |
| Capture ratios | Compare fund and benchmark returns within separately defined benchmark-up and benchmark-down periods. | Specify arithmetic/geometric conventions, compounded windows and treatment of zero benchmarks; no universal capture target proves suitability. |
Original measurement exercise: Three equally weighted periodic returns are +2%, −1%, +3%, with a 0% target. Downside deviation under the table’s convention is sqrt((0+1+0)/3) = 0.57735 percentage points. Mean return is 1.33333%; Sortino is about 2.309 for this tiny unannualized sample. Omega is (2+3)/1 = 5. Neither statistic establishes low future tail risk.
A tail-loss estimate is not a stop-loss guarantee
For an original historical exercise, take 100 equally weighted daily loss observations and use the explicit nearest-rank convention. Sort losses from smallest to largest; 95% VaR is observation 95. If that value is $20,000 and the five larger losses are $25,000, $30,000, $40,000, $60,000 and $100,000, their average is $51,000. That five-observation tail average illustrates severity beyond the selected quantile under this convention. It is not the exact expected-shortfall convention for every finite sample or distribution.
Variance-covariance approaches may impose normality and linear approximations; historical simulation uses empirical scenarios; Monte Carlo depends on the simulated model and inputs. It is inaccurate to say every VaR method requires a normal distribution. More simulations do not create independent historical evidence or ensure that every relevant crisis was modeled.
Your turn: Why can a current-portfolio shock using a past crisis still be useful, despite Chapter 13’s criticism?
Open the explained answer
It is a conditional scenario: how today’s positions might respond to specified historical-like shocks. It is not a claim that today’s holdings were owned then or that the past crisis will recur exactly. State repricing assumptions, instruments unavailable historically, liquidity and financing responses. Combine it with forward-looking and reverse stresses instead of discarding or overinterpreting it.
Due diligence and lifecycle
14. Use questions to request evidence, not reassurance
Read: Chapter 13 quantitative and qualitative sections, pp. 221–237. Goal: Combine track-record analysis with a verifiable process review.
The chapter recommends combining statistics with detailed questioning. That combination is useful when the questions produce records that can contradict a polished explanation. Credentials, prestigious employers, referrals and coinvestment can be relevant background; they do not certify competence, independence or integrity.
Compare a manager to an appropriate peer set and examine rolling windows, attribution and changed AUM. A result generated at small size, with a different team or on paper may not transfer to the current proposed product. Good performance across several names still can depend on one shared factor.
A tail observation should not be deleted merely because it looks unusual. Investigate whether it is a data error or an economically relevant event, and report the effect of any justified correction. Current-portfolio stress scenarios and observed histories answer different questions; both can contribute if honestly labeled.
Original worked example
A fictional fund reports a 10% year. Attribution shows +9 percentage points from one event position and +1 point from the rest of the portfolio. The 10% total is correct, but the concentration of contribution creates a different sustainability question from a broad record of repeatable contributions. A single large success neither proves nor disproves skill.
Your turn: Write three questions linking the number to evidence. What must be verified before attributing a predecessor fund’s performance to a new manager?
Open the explained answer
Request dated position and contribution records, the original event thesis and decision authority, and the behavior of the rest of the strategy across conditions. For a predecessor record, verify who made and controlled the decisions, team composition, fees, mandate, operational arrangements and the basis for any claimed portability. Distinguish reported, audited and independently reconciled records.
A convincing conversation ends with a checkable record. Keep unresolved evidence gaps visible.
15. Operational diligence follows cash, authority and prices
Read: Chapter 13 operational, background and ongoing sections, pp. 238–248. Goal: Design controls that can be tested independently.
The operational sections examine money movement, provider roles, allocation, valuation, reporting and continuity. The central question is who can act, who checks the action and where the evidence is kept. A list of well-known providers is not sufficient; verify the relevant relationship through independently established contact channels.
Some statements in the chapter are prescriptive preferences, not universal legal requirements. Side pockets, side letters, error compensation and differing account terms require analysis of actual disclosures and agreements. They are not automatically identical across funds or investors.
Background findings require identity resolution, current status and scope. A complaint is different from a finding, a dismissed case from a conviction, and no record found from proof that no event ever occurred. Use qualified, lawful, authorized processes; the appendix’s historical use of sensitive records is not permission to collect or publish personal information.
Original worked example
A fictional payment workflow lets the portfolio manager create a payee, approve the transfer and reconcile the bank statement. Three steps exist, but one person controls all of them. An independent approver with verified bank details and separate reconciliation would address different failure points; two nominal signatures are weak if they share the same access or simply rubber-stamp.
Your turn: Choose one valuation control, one cash control and one continuity test. For each, state the evidence, responsible owner and review trigger.
Open the explained answer
Examples: independently supported prices with exception logs owned by valuation oversight; independently confirmed payees and dual authorization owned by treasury/operations; a documented recovery exercise owned by a named continuity officer. Record dates, failures and remediation. Auditor or administrator names alone do not prove these controls operated. Ongoing diligence should revisit personnel, capacity, exposures, reporting revisions and provider changes.
Test how the control works. A policy document is weaker evidence than a reconciled, reviewed record.
16. Distinguish a weak hypothesis from a struggling business
Read: Chapter 14, pp. 249–264. Goal: Investigate alpha decay without selecting favorable endpoints.
The chapter uses hypothetical hedged size, momentum and accrual portfolios to show that attractive long-term numbers can coexist with difficult stretches. These examples are historical constructions, not audited hedge fund returns and not new results produced by this workbook. Publication dates, data availability, hedge mismatch and fees matter.
In a fresh study, use point-in-time membership and financial-statement availability, delisting outcomes, realistic borrowing and trading costs, and a stated funding convention. A long-minus-short return plus a cash rebate is not a complete financed portfolio unless notionals, collateral, fees and rebalancing are specified. An accrual signal also needs a precise accounting definition; a high accrual is not proof of fraud.
Distinguish arithmetic averages from compound growth. Some surrounding lifecycle prose does not consistently describe the table statistics; calculate from the series and clearly label the metric. Similarly, do not rely on a plotted scale when the table contains values outside that scale. This companion does not validate the chapter’s reported performance.
Original worked example
Our fictional two-period strategy earns +50% then −40%. Arithmetic average return is +5%, but wealth moves from 100 to 150 to 90. The two-period CAGR is sqrt(0.9)−1 = −5.13% per period. The profitable-looking average does not describe investor wealth. A business can also face redemptions before a later recovery, which is a separate path risk.
Your turn: Prepare three competing explanations for recent underperformance: edge erosion, implementation failure and cyclical exposure. What evidence would help distinguish them? Should you automatically increase leverage because a historical average was favorable?
Open the explained answer
Edge erosion calls for evidence of reduced gross opportunity or crowding; implementation failure calls for changed costs, borrow, fill quality or capacity; cyclicality calls for a predeclared factor or regime relation with evidence beyond a convenient story. More leverage magnifies model and path risks and cannot establish which explanation is right. Use fixed review dates, thresholds and escalation authority rather than tuning a response to the latest chart.
Diagnose the mechanism before changing the size. Underperformance is an observation, not its own explanation.
17. Turn seven forecasts into an audit plan
Read: Chapter 15, pp. 265–280. Goal: Distinguish a historical forecast from an observed trend.
The final chapter proposes seven directions for industry development. Its rationale links competition, institutions, crisis experience and disclosure demands. The workbook task is to define what evidence would count for or against each prediction. It is not to assume the forecast was correct because it sounds plausible now.
Compare like products and investor classes. Changing sample composition can create a falling fee average even when every continuing fund charges the same fee. “More liquidity” may mean a shorter notice period, a different wrapper or better actual redemption outcomes; those are distinct measurements.
Replication also needs a target. Reproducing a category’s common risk exposures does not reproduce an individual manager’s trades, discretionary judgment or idiosyncratic skill. A manager should explain what service or exposure remains after accounting for an accessible, appropriately matched alternative.
Original worked example
A fictional database has five continuing funds charging 2% management fees. Five new funds join at 1%. The equal-weight cross-sectional mean falls from 2% to 1.5%, while the continuing cohort’s fee stays 2%. Both statements are true; only one measures fee changes within the original funds.
Your turn: For the seven forecasts below, choose one metric, comparison date and falsifying observation each. Why should you save the original forecast wording and sample rule before researching outcomes?
Open the explained answer
It prevents redefining success after seeing the data. For example, a lower headline fee with higher pass-through expenses may not mean a lower all-in cost. Preserve unchanged definitions, document missing or closed funds, compare continuing and new cohorts separately and record mixed results. This workbook supplies an audit design, not a current verdict on all seven forecasts.
A forecast needs an observable test. Plausibility and hindsight are not the same as evidence.
| Source forecast | Original measurement task | A result that would weaken a broad interpretation |
|---|---|---|
| Lower fees | Track all-in costs for a fixed cohort and equivalent investor class, alongside changing cross-sections. | Headline fees fall while pass-through costs keep total cost unchanged or higher. |
| Increased liquidity | Compare notice periods, lockups, gates and actual access in comparable products and stresses. | Shorter stated windows coexist with frequent suspension or assets that remain hard to realize. |
| Increased transparency | Track independently supported position/factor data, reporting delay and valuation detail. | More pages of marketing material without better usable or verifiable information. |
| Emerging managers | Define manager versus fund age and include closed and unreported funds where possible. | Apparent superiority disappears after survival, backfill, capacity and risk adjustments. |
| Niche strategies | Specify the niche, scalable opportunity, costs and competitive response. | Novelty produces thin evidence or high costs rather than persistent risk-adjusted value. |
| Limited replication | Define the target category, investable comparator, fees and out-of-sample tracking. | Performance depends on a fitted historical model that fails on new observations. |
| Hedge fund/private equity convergence | Distinguish firm-level product expansion from convergence in underlying liquidity and mandates. | The same brand offers both products, but the cash flows, control and investor terms remain materially different. |
For current disclosure research, Form 13F is only a partial, delayed view of covered reportable positions, not a complete trading book. It does not report short equity or short option positions. Do not infer a manager’s current net exposure from it. Consult the SEC’s Form 13F FAQ rather than carrying forward the book’s simplified reporting description.
Three appendix assignments
18. Update a resource by function, not by copying an old bookmark
Read: Appendix 1, pp. 283–292. Goal: Create a source hierarchy and a point-in-time data log.
The bookmarks are explicitly dated November 2008. A familiar domain can change purpose, move or disappear. Preserve the research function—filings, valuation, events, macro series, risk evidence—rather than assuming every old link is still authoritative. Search-result visibility or a paid platform does not certify accuracy.
Use primary documents when the task is to establish reported facts, and use commentary as a hypothesis source. For a new empirical test, distinguish the observation period from the release date and later revisions. SEC EDGAR is a starting point for public filings; economic-data research also needs vintage information when available, as illustrated by the St. Louis Fed’s ALFRED vintage-data guidance.
Original worked example
A fictional economic observation refers to March, was released in April and revised in June. A decision on March 31 could not use the April release or the June revision. A backtest using the final revised value at March 31 would include future information.
Your turn: Build five resource rows: company filings, industry information, macro data, execution data and professional/compliance checks. Give each a purpose, provenance requirement, timestamp and known blind spot.
Open the explained answer
Example: a company filing row records the issuer, official filing, acceptance date, period and amendments, and notes that it is reported information rather than independent verification of every claim. A macro row records release/vintage dates; an execution row requires actual trade/order data for fill questions; a professional-check row records entity identity, jurisdiction, scope and check date. Do not silently replace historical timestamps with today’s revised data.
Save what was knowable when. A modern data source does not automatically provide a point-in-time history.
19. Choose reading to resolve a specific weakness
Read: Appendix 2, pp. 293–297. Goal: Turn a reading list into a purposeful learning sequence.
The appendix’s reading suggestions reflect contributor judgments at publication. A famous book can be valuable without providing a tested method, and an engaging interview can be informative without a representative sample. Do not carry forward the appendix’s used-book prices or treat old editions as current legal guidance.
Organize reading by the gap it closes: accounting, pricing, execution, behavioral controls, risk or case analysis. After reading, produce a small falsifiable calculation or a concrete change to your research process. Inspiration is not validation.
Original worked example
A learner can compute technical indicators but cannot reconcile net income to operating cash flow. The immediate learning task is a cash-flow bridge using financial statements, not adding another chart rule. Another learner understands valuation but cannot measure filled and unfilled order costs; that learner needs an execution exercise.
Your turn: Select one source-listed accounting work, one risk/case work and one behavioral work available to you. For each, write a question, an output and a limit. You do not need to buy anything for this assignment.
Open the explained answer
A suitable plan might produce an earnings-to-cash reconciliation, a funding-stress timeline and a pre-mortem checklist. Record what each source can establish and what it cannot. A case history cannot supply a representative expected return; a behavioral explanation cannot identify motives from a chart. Judge completion by the output rather than by pages read.
Read to solve a defined problem. A longer shelf is not a stronger process.
20. Read a background report without turning allegations into findings
Read: Appendix 3, pp. 299–311. Goal: Distinguish identity, scope, procedural status and relevance.
The supplied report uses a fictitious subject and draws on case material. It is a teaching sample, not a current report on an actual person to be investigated. We do not reproduce its personal details or reuse its subject’s name as a real allegation.
Professional verification requires accurate identity matching and lawful access. An allegation, disposition, sanction and unresolved identity match belong in different fields. “No records found” is a statement about a search’s scope and date, not a guarantee of a clean lifetime history. A final investment decision should connect material verified facts to the role and mandate.
Original worked example
In our different fictional case, a report finds a civil filing against an entity with a similar name, a confirmed professional registration and an unverified biography claim. The registration does not resolve the lawsuit identity or verify the biography. Each item needs its own provenance and status.
Your turn: Create a four-column evidence record for those three items: identity confidence, primary document needed, present status and decision relevance. What should be done before making an adverse factual statement about a person?
Open the explained answer
Resolve the identity using appropriate nonpublic handling of authorized identifiers, obtain the actual filing and disposition where lawful, distinguish claims from findings, and give uncertain facts an uncertainty label. Registration confirms only the relevant registration information and date; it does not prove skill or approval of an investment. Seek an appropriate professional review for sensitive or material background findings, rather than publishing an unresolved allegation.
Be precise about what was checked and found. Evidence quality includes procedural fairness and current status.
Three reproducible research labs
The assignments below are original protocols, not completed tests or forecasts. Keep the book as context, retain all trials and use only data you are entitled to analyze. A spreadsheet or notebook is sufficient for the manual labs; the indicator supports the chart-proxy comparison.
| Lab | Fixed rules and variation | Outputs and interpretation |
|---|---|---|
| A. Execution benchmarks | Select ten complete historical sessions for one liquid listed equity on standard five-minute bars. Fix timezone, session and data feed. Compute the close-volume VWAP proxy and equal-bar TWAP proxy. Repeat the same sessions on one-minute bars; change only bar size. | Record end-window values and gaps in bps, missing bars/volume and differences by bar size. Differences show sampling sensitivity. They do not prove which broker could obtain a better fill. Do not add an assumed fill or call the gap alpha. |
| B. Alpha model sensitivity | Use the same 60 aligned monthly observations for a specified return series and benchmark, with matched risk-free observations and a stated net/gross basis. Compare a constant-mean excess-return model with a one-factor excess-return regression; then, if justified, preselect one additional factor and document why. | Record intercept, coefficients, uncertainty intervals, residual distribution and changes in conclusions. Do not select factors until the alpha becomes positive. Serial dependence and small samples can undermine conventional standard errors; state the estimation method and limits. |
| C. Strategy lifecycle and costs | Use two fictional 50-trade ledgers: A has 20 gains of $300 and 30 losses of $100; B has 40 gains of $50 and 10 losses of $300. Compare average all-in costs of $5, $15 and $30 per trade while keeping outcome counts and gross payoffs fixed. These are assumed ledgers, not market data or estimated probabilities. Specify how real point-in-time records would replace the assumptions. | Record gross and net P&L, observation count, assumptions and cost sensitivity. Deterministic assumed counts are not a Monte Carlo result or a validated edge. In a live-data study, include all eligible events, delistings, borrow, financial-statement lags and a reserved evaluation period. |
Check the interpretation of the labs
Lab A measures chart aggregation and weighting, not implementation shortfall; actual shortfall requires order decisions, quantities and fills. Lab B asks whether a residual is robust to a defensible model, not whether model searching can manufacture one. Lab C: A has $3,000 gross profit and net profits of $2,750, $2,250 and $1,500 at the three cost levels. B loses $1,000 gross and $1,250, $1,750 and $2,500 net. It isolates arithmetic sensitivity; it cannot establish actual probabilities or strategy lifespan. Record unfavorable and inconclusive outcomes alongside favorable ones.
Pine indicator: execution benchmark lens
This original Pine v6 indicator implements a narrow Chapter 8 teaching exercise. It compares two averages of completed chart bars within a chosen same-day window. It does not reconstruct the book’s trading examples or execute an algorithm. The source supplies the execution concepts; our use of bar closes, review threshold and state definitions is a simplified educational implementation.
What This Indicator Shows
On a standard intraday minute chart, it accumulates a close-volume VWAP proxy, Σ(close×bar volume)/Σ(bar volume), and an equal-bar TWAP proxy, Σ(close)/completed-bar count. For equal-duration, continuously available bars, the second is a uniformly sampled time average. Both are approximations: bar-close weighting is not transaction-level VWAP, and one sample per bar is not continuous-time TWAP.
The default window is 09:30–16:00 in the symbol’s exchange timezone, an illustrative U.S. equity window. Use a same-day session whose boundaries align with the chart bars; use complete standard minute bars and check holidays, early closes and feed coverage. This baseline does not implement an exchange calendar or overnight sessions. It resets on the first valid completed in-window bar belonging to a new daily bucket. If loaded history starts mid-window, the first window is partial and must be excluded from comparisons.
The Data Window shows the signed VWAP-minus-TWAP gap in basis points and sampled-bar count. One REVIEW diamond appears at the first completed bar whose absolute gap reaches the chosen threshold in the window. The default 20 bps is our research prompt, not a trade recommendation. The review state stays active for the rest of that window; falling back below the threshold does not re-arm it. The next day’s first valid completed sample resets the state. There are no entries, pending orders, replacement setups or exit/expiry instructions.
What This Indicator Does Not Show
It cannot identify bid/ask quotes, hidden orders, venue liquidity, queue priority, institutional intent or actual broker fills from OHLCV. It does not calculate a trader’s true implementation shortfall because no decision price, intended quantity, unfilled balance or actual executions are supplied. It does not measure fund alpha, return, profitability or best-execution compliance. A close crossing either line is not a defined trading signal.
All accumulated observations and review events use confirmed bars; there is no higher-timeframe request or future-data lookup. During a live unfinished bar, values retain the last completed sample for the active day. No values display outside the selected window or before the first valid sample of a new day. Revised history, session settings, symbol, adjustment or bar-size changes can change the results. See TradingView’s bar-state documentation (paid link) for confirmed-bar behavior.
Missing or nonpositive volume invalidates the VWAP proxy and gap for the rest of that daily window rather than silently treating incomplete weights as complete. TWAP still uses valid positive closes. Missing bars are not synthesized; they can make the sampled average incomplete without being detected by the script. Tick-volume instruments require separate interpretation and are not interchangeable with traded-share volume.
Validation status
Static code review, arithmetic examples and exact displayed-code round-trip checks have been completed. The indicator has not been compiled in TradingView, applied to a live chart or browser-tested for copying. No fills, win rates, returns or execution advantage are claimed. Before using alerts, compile it and manually verify a full window’s bar count, weighted sums, reset and first-trigger behavior. Set alerts once per bar close.
View Pine Script Indicator Code
//@version=6
indicator("CandleChasers Execution Benchmark Lens", overlay=true)
// Original chart-bar proxies. No orders and no reconstruction of actual trades.
sessionWindow = input.session("0930-1600", "Same-day window (exchange timezone)")
reviewBps = input.float(20.0, "VWAP/TWAP gap review threshold (bps)", minval=0.1, step=1.0)
if not timeframe.isintraday or not timeframe.isminutes
runtime.error("Use a standard intraday minute chart with equal-duration bars.")
bool inWindow = not na(time(timeframe.period, sessionWindow))
int calendarDay = time("D")
var int activeDay = na
var float sumPrices = 0.0
var int countBars = 0
var float sumPV = 0.0
var float sumVolume = 0.0
var bool volumeComplete = true
var bool reviewActive = false
var float twapProxy = na
var float vwapProxy = na
var float gapBps = na
bool reviewEvent = false
if barstate.isconfirmed and inWindow and not na(close) and close > 0
bool resetWindow = na(activeDay) or calendarDay != activeDay
if resetWindow
activeDay := calendarDay
sumPrices := 0.0
countBars := 0
sumPV := 0.0
sumVolume := 0.0
volumeComplete := true
reviewActive := false
sumPrices += close
countBars += 1
if na(volume) or volume <= 0
volumeComplete := false
else
sumPV += close * volume
sumVolume += volume
twapProxy := sumPrices / countBars
vwapProxy := volumeComplete and sumVolume > 0 ? sumPV / sumVolume : na
gapBps := not na(vwapProxy) and twapProxy > 0 ? 10000.0 * (vwapProxy / twapProxy - 1.0) : na
if not na(gapBps) and math.abs(gapBps) >= reviewBps and not reviewActive
reviewEvent := true
reviewActive := true
bool showValues = inWindow and not na(activeDay) and activeDay == calendarDay
plot(showValues ? vwapProxy : na, "Close-volume VWAP proxy", color=color.teal, linewidth=2, style=plot.style_linebr)
plot(showValues ? twapProxy : na, "Equal-bar TWAP proxy", color=color.orange, linewidth=2, style=plot.style_linebr)
plot(showValues ? gapBps : na, "VWAP minus TWAP (basis points)", display=display.data_window)
plot(showValues ? countBars : na, "Completed sampled bars", display=display.data_window)
plotshape(reviewEvent, title="First benchmark-gap review", style=shape.diamond, location=location.abovebar, color=color.purple, size=size.tiny, text="REVIEW")
alertcondition(reviewEvent, title="Benchmark gap review", message="Confirmed chart-bar VWAP/TWAP proxy gap reached the review threshold. This is not a trade signal or fill.")
Hand-check the weighting and review state
| Completed samples in the same window | Expected output |
|---|---|
| Close 100, volume 100; then close 102, volume 300 | TWAP = 101. VWAP = (10,000+30,600)/400 = 101.5. Gap = (101.5/101−1)×10,000 ≈ +49.50 bps; first REVIEW at 20 bps. |
| Then close 100, volume 100 | TWAP ≈ 100.6667; VWAP = 101.2; gap ≈ +52.98 bps. No second REVIEW in this window. |
| Next day, first valid close 90 with volume 200 | Both proxies equal 90; gap 0; count 1; review state reset. |
| Next completed bar has missing volume | TWAP can update from its valid close; VWAP and gap become unavailable for the rest of that daily window. No gap event can be generated while unavailable. |
Your turn: If a gap narrows below 20 bps and then exceeds it again in the same window, should another REVIEW occur? Why can the one-minute and five-minute end values differ?
Open the explained answer
No second REVIEW occurs because the first event latches the window’s state. Bar-close prices and volume allocations change with aggregation; a five-minute close does not preserve every one-minute price×volume contribution. TWAP sampling changes too. Neither output is guaranteed to equal a transaction-level benchmark or an executable fill.
Four reusable worksheets
Reuse these field sets in your notebook. Keep the original entry intact when you revise a hypothesis, and date each new version. The decision sheet is a fictional research aid, not a live order form.
| Field | Record |
|---|---|
| Source and claim | Printed page or primary record; who made the claim and at what date. |
| Evidence status | Definition, source opinion, reported observation, independently checked record or our assumption. |
| Availability timestamp | When the information could actually have been known; include later revisions separately. |
| Unit and scope | Currency, frequency, gross/net basis, universe, instrument and time horizon. |
| Alternative and falsifier | A competing explanation and a specific observation that would weaken your preferred one. |
| Field | Record |
|---|---|
| Purpose | Portfolio role, authority and reason for the proposed exposure. |
| Mechanism | What is bought/sold, why its payoff should differ from a fair benchmark, and timing. |
| Risk decomposition | Directional/factor, option, credit, currency, financing, counterparty and liquidity risks. |
| Implementation | Quantity, capacity, price assumptions, borrow, actual contract terms and nonexecution handling. |
| Economics | Success/failure branches, costs, fee bases and sensitivity to uncertain inputs. |
| Decision | Approve/defer/reject for the fictional case; missing evidence and change-of-mind conditions. |
| Field | Record |
|---|---|
| Hypothesis | One falsifiable claim, declared before searching for the best result. |
| Dataset | Source, lawful access, dates, sampling, adjustments, vintage and missing values. |
| Experiment | Fixed baseline, changed parameter, tested values and reserved evaluation period. |
| Metrics | Numerator, denominator, horizon, units and uncertainty method. |
| Bias controls | Selection, backfill, look-ahead, dependence, costs, delistings and unreported trials. |
| Interpretation | What supports, weakens or leaves the claim unresolved; include failures. |
| Field | Record |
|---|---|
| Prior prediction | The dated claim and probabilities or scenario assumptions before the outcome. |
| Observed evidence | What changed and when it became available. |
| Attribution | Separate hypothesis, implementation, exposure and business explanations. |
| Control / action | Named responsible owner, authority, escalation path and documented response. |
| Next review | Date or defined trigger; one process improvement and one remaining uncertainty. |
Capstone: test an alpha pitch from idea to investor outcome
Everything in this case is fictional. A prospective allocation is described as “market neutral, portable alpha with strong execution.” It has $10 million NAV, $12 million long equities at estimated beta 1.1 and $8 million short equities at beta 0.9. It reports an 11% gross annual return; the same-period risk-free reference is 3% and the chosen market returned 8%. There are only twelve monthly observations and no independent confirmation of valuations or fees.
For the positive-year fee exercise only, assume 2% management fees on opening NAV and 20% incentive fees on the profit left after management fees, no prior high-water-mark adjustment and no other expenses. For liquidity, assume $1 million cash, $3 million of investor obligations due in 30 days and $4 million of assets expected to become accessible in 90 days, subject to actual fund terms.
- Calculate simple gross and net exposure, then estimated beta-adjusted net exposure.
- Calculate a single-period CAPM residual from the supplied beta exposure. Explain why it is not proof of skill.
- Compute the simplified net investor return after the stated fees.
- Stress the portfolio: long stocks fall 10%, while the shorted stocks rise 5%. State dollar and NAV losses before costs.
- Calculate the 30-day cash gap and identify the evidence needed to assess realistic payment capacity.
- Evaluate the execution pitch using the filled/unfilled cost example in lesson 8. Explain why a chart VWAP comparison is insufficient.
- Produce a two-page decision memo plus the four worksheets. Identify five verification tasks, a behavioral failure mode, an operational control and the authority to approve any allocation.
Open the capstone worked answer
Gross exposure is ($12m+$8m)/$10m = 200%; simple net is 40%. Beta-adjusted dollar exposure is $12m×1.1−$8m×0.9 = $6m, or beta 0.6 relative to NAV under the simple linear model. Dollar net, beta and total risk are different. The “market neutral” label is unsupported by the stated beta.
Expected CAPM return is 3%+0.6×(8%−3%) = 6%; the supplied gross-return residual is 5 percentage points. It is a single-period model illustration with only a short, unverified record and incomplete factors. A net-of-fee calculation gives a different residual and neither establishes statistically reliable, persistent skill.
Starting with $100, 11% gross produces $111; management fees remove $2, then incentive fees remove 20% of $9 = $1.80. Ending investor value is $107.20, or 7.2% net under our convention. On that same illustrative expected return, the net residual is 1.2 points. Costs and actual terms may change it.
The stress loses $1.2m on longs and $0.4m on shorts, totaling $1.6m, or 16% of opening NAV before financing, borrow, fees and forced-liquidation effects. The 30-day cash gap is $2m; expected 90-day access does not fill it. Verify notice, gates, suspension, cash obligations, realizable assets and financing rather than assuming a scheduled date is guaranteed.
Execution evidence needs decision timestamps, intended quantity, route/order changes, actual fill reports, unfilled balances and explicit costs. The lesson 8 example’s shortfall is $2,230, including the unfilled opportunity component, or 44.6 bps of intended notional. A chart-bar average cannot substitute for those records.
Deferring for verification is defensible. Tasks include independently confirmed performance and valuation, actual fees and legal terms, provider relationships and cash controls, look-through exposures/financing, and real execution records. Challenge confirmation bias with a documented independent review. A multi-person payment-control workflow and reconciled statements should be tested in operation. State the authorized decision maker and evidence that could change the decision; do not invent favorable facts to approve.
| Criterion | Points | Completion standard |
|---|---|---|
| Exposure and model | 25 | 200% gross, 40% net, beta 0.6 and 5-point gross residual, with assumptions. |
| Fee and stress arithmetic | 25 | 7.2% net investor return and 16% stress loss, correctly labeled. |
| Liquidity and implementation | 20 | $2m cash gap; delayed access and filled/unfilled execution evidence addressed. |
| Evidence and controls | 20 | Five actionable verification tasks, a bias-control pair and an operational control. |
| Decision and uncertainty | 10 | Reasoned authority, decision, monitoring plan and change-of-mind evidence. |
Completion check: Attempt all twenty lessons, the derivative drill, measurement exercises and seven forecast rows before checking answers. Complete each worksheet, specify all three lab protocols without inventing outcomes and score at least 80/100 on the capstone. Rework any Kelly, leverage, liquidity or VaR misunderstanding even if the total score passes. This is a learning standard, not a qualification to manage money.
Final Takeaway
A credible alpha claim needs a chain of evidence: an appropriate model, a lawful and timely research process, executable economics, robust risk measurement and operational controls. The strongest result of studying this book is a better ability to find a broken link in that chain—and to say clearly what still needs to be verified.
Risk Disclosure
This material is educational and is not personalized investment, legal or tax advice. Trading and investing can cause substantial losses. Leverage, short selling and derivatives can amplify losses, and some exposures can lose more than the capital initially committed. Private funds and structured products may have limited liquidity, valuation uncertainty and counterparty risk. Historical, hypothetical and model-based results do not predict future performance. Examples use simplified assumptions and omit costs unless stated. The Pine indicator is an uncompiled educational proxy and does not establish fills, execution quality or profitability. Verify current rules, actual documents and suitability with appropriately qualified professionals before making real decisions.
This is an independent educational companion. CandleChasers is not affiliated with or endorsed by the editor, contributors, publisher, firms or individuals discussed.