Imported from arjundrath-star/klade-analyst (clients/_archived/reid/skills/benchmark-construction/SKILL.md). Install upstream with npx skills add arjundrath-star/klade-analyst --skill benchmark-construction. Copyright stays with the author.

benchmark-construction

Construct a custom performance benchmark tailored to a specific portfolio's mandate, style, and constraints. A bad benchmark makes every performance number meaningless — if you're measuring a small-cap value fund against the S&P 500, you're measuring noise, not skill. This skill builds proper benchmarks from index constituents, blended indices, factor exposures, or peer group returns, following CFA Institute / GIPS standards.

Trigger

• "Build a benchmark for [portfolio/fund/strategy]"
• "Custom benchmark for [mandate]"
• "Blended benchmark using [indices]"
• "Peer group benchmark for [strategy type]"
• "Is [index] the right benchmark?"
• "Benchmark appropriateness review"

Inputs

Required

• Portfolio description: mandate, strategy type, asset classes, constraints
• Holdings or allocation: current portfolio composition (even approximate)

Optional

• Candidate benchmarks: indices to evaluate or blend (default: skill selects based on mandate)
• Peer group: list of comparable funds/strategies for peer-based benchmark
• Weighting scheme: market-cap, equal-weight, fundamental-weight (default: market-cap)
• Rebalancing frequency: monthly / quarterly / annual (default: quarterly)
• Currency: base currency for multi-asset benchmarks (default: USD)
• Custom constraints: ESG exclusions, sector caps, geography limits

Dependencies

• financial-data-api — data source stack (see ../financial-data-api/SKILL.md)

⚠️ DATA SOURCING MANDATE (NON-NEGOTIABLE)

1. Massive.com API (PRIMARY for index and constituent prices):

   • web_fetch("https://api.massive.com/v2/aggs/ticker/{TICKER}/range/1/day/{FROM}/{TO}?adjusted=true&apiKey=${MASSIVE_API_KEY}") for historical prices
   • Use for all return calculations on benchmark components

2. FRED API (PRIMARY for macro benchmarks and rates):

   • Treasury indices, inflation data, credit spread indices
   • web_fetch("https://api.stlouisfed.org/fred/series/observations?series_id={ID}&api_key=${FRED_API_KEY}&file_type=json")
   • Key series: DGS10, DGS2, DTB3, BAMLH0A0HYM2, BAMLC0A0CM, CPIAUCSL

3. SEC EDGAR (for fundamental-weighted benchmarks):

   • Revenue, earnings, book value for fundamental weighting schemes
   • CIK lookup + XBRL extraction per standard protocol

4. web_search (SUPPLEMENTARY):

   • Index methodology documents, constituent lists, fund fact sheets
   • Historical index returns when direct calculation isn't feasible
   • NEVER use for price data that Massive.com provides

Methodology

Step 1: Assess the Portfolio Mandate

Before constructing anything, characterize the portfolio along these dimensions:

CFA Institute benchmark properties (ALL must be satisfied):

1. Specified in advance — defined before the measurement period
2. Appropriate — reflects the manager's style and universe
3. Measurable — returns can be calculated on a timely basis
4. Unambiguous — constituents and weights are clearly defined
5. Reflective of manager's knowledge — manager knows and could hold the constituents
6. Investable — a passive alternative exists or could be constructed
7. Owned — manager accepts it as the standard for comparison

Step 2: Identify Candidate Benchmarks

Based on the mandate assessment, identify 3-5 candidate benchmarks:

web_search("[asset class] [geography] [style] index benchmark [provider]")
web_search("[strategy type] standard benchmark institutional")
web_search("[index name] methodology constituents factsheet")

Common benchmark families:

Step 3: Evaluate Benchmark Fit

For each candidate, calculate fit metrics:

Holdings overlap:

Overlap = Σ min(w_portfolio_i, w_benchmark_i) for all securities i

Target: >60% overlap for the benchmark to be "appropriate."

Style/factor alignment:

• Compute portfolio vs benchmark factor exposures (market beta, size, value, momentum)
• Large factor mismatches indicate the benchmark doesn't reflect the portfolio's style

Tracking error analysis:

Historical TE = std(R_portfolio - R_benchmark) × √252

Expected TE should be consistent with the manager's expected alpha and risk budget.

Return correlation:

Correlation = corr(R_portfolio, R_benchmark)

Target: >0.90 for the benchmark to be representative.

Step 4: Construct the Benchmark

Option A: Single Index

• Select the best-fit index from Step 3
• Document why alternatives were rejected

Option B: Blended Benchmark When no single index fits, create a weighted blend:

R_benchmark = w_1 × R_index1 + w_2 × R_index2 + ... + w_n × R_indexn

Blending rules:

• Weights must sum to 100%
• Each component must have ≥5% weight (below that, it's noise)
• Maximum 5 components (beyond that, the benchmark itself becomes opaque)
• Weights rebalanced at specified frequency (quarterly default)
• Document the rationale for each weight

Example: Balanced fund benchmark:

60% S&P 500 + 30% Bloomberg US Aggregate + 10% MSCI EAFE = Balanced 60/30/10

Option C: Peer Group Benchmark When constructing from peer returns:

• Identify universe of comparable funds/strategies (minimum 20 for statistical validity)
• Calculate equal-weight or AUM-weight average return of the peer group
• Show distribution: 25th, 50th, 75th percentile
• Limitation: peer groups have survivorship bias (failed funds drop out) — document this

Option D: Factor-Based / Custom Benchmark When the mandate has unique constraints (ESG, sector caps):

• Start with the closest standard index
• Apply the portfolio's constraints to the index constituents
• Reweight remaining constituents per the chosen scheme
• Rebalance at specified frequency
• Document every constraint applied

Step 5: Build Historical Return Series

For the chosen benchmark construction, compute historical returns:

1. Pull adjusted daily prices for all constituents from Massive.com
2. Apply the weighting scheme at each rebalance date
3. Calculate daily benchmark returns between rebalances
4. At rebalance dates, update weights per methodology
5. Chain-link daily returns into a cumulative NAV series

NAV_t = NAV_{t-1} × (1 + R_benchmark_t)
R_benchmark_t = Σ(w_i,t × R_i,t) for all constituents i

Step 6: Validate the Benchmark

Run validation checks:

Statistical validation:

• Tracking error vs portfolio: is it in the expected range?
• Correlation with portfolio: >0.90?
• Information ratio: does excess return look like alpha or benchmark mismatch?

Practical validation:

• Is the benchmark investable? Could you buy an ETF that tracks it?
• Is the benchmark transparent? Can returns be independently verified?
• Does the manager agree this is a fair comparison?

Gaming detection:

• Is the benchmark chosen to make the manager look good? (e.g., small-cap fund benchmarked to large-cap index)
• Does the benchmark have a known structural beta mismatch?
• Is the blend overly complex to obscure poor performance?

Step 7: Document Benchmark Specification

Create a formal benchmark specification document suitable for IPS (Investment Policy Statement) inclusion.

Output Format

📐 Benchmark Construction — [Portfolio/Strategy Name]
Prepared: [Date] | Base Currency: [CCY] | Rebalancing: [Frequency]

━━━ PORTFOLIO MANDATE ASSESSMENT ━━━
| Dimension      | Portfolio Characteristic      |
|----------------|------------------------------|
| Asset Class    | [e.g., US Equity]            |
| Geography      | [e.g., Domestic]             |
| Market Cap     | [e.g., Large-cap]            |
| Style          | [e.g., Growth]               |
| Sector Focus   | [e.g., Broad / Technology]   |
| Constraints    | [e.g., ESG exclusions]       |
| Income Target  | [e.g., Total return]         |

━━━ CANDIDATE BENCHMARK EVALUATION ━━━
| Benchmark         | Ticker | Overlap | Correlation | Tracking Error | Fit Score |
|-------------------|--------|---------|-------------|----------------|-----------|
| [Index 1]         | XXX    | XX.X%   | 0.XX        | XX.X%          | [A/B/C]   |
| [Index 2]         | XXX    | XX.X%   | 0.XX        | XX.X%          | [A/B/C]   |
| [Index 3]         | XXX    | XX.X%   | 0.XX        | XX.X%          | [A/B/C]   |
| [Blended Custom]  | —      | XX.X%   | 0.XX        | XX.X%          | [A/B/C]   |

Fit scoring: A = excellent (overlap >70%, corr >0.95) | B = acceptable | C = poor fit

━━━ SELECTED BENCHMARK ━━━
Type: [Single Index / Blended / Peer Group / Custom]

[If blended:]
| Component        | Ticker | Weight | Rationale                        |
|------------------|--------|--------|----------------------------------|
| [Index 1]        | XXX    | XX%    | [Core equity exposure]           |
| [Index 2]        | XXX    | XX%    | [Fixed income allocation match]  |
| [Index 3]        | XXX    | XX%    | [International sleeve]           |
| **Total**        |        | 100%   |                                  |

Rebalancing: [Quarterly on last business day]
Reconstitution: [Per underlying index methodology]

━━━ CFA BENCHMARK PROPERTIES CHECKLIST ━━━
| Property               | Status | Notes                              |
|------------------------|--------|------------------------------------|
| Specified in advance   | ✅/⚠️  | [Defined as of YYYY-MM-DD]        |
| Appropriate            | ✅/⚠️  | [Style and universe match]         |
| Measurable             | ✅/⚠️  | [Daily returns available]          |
| Unambiguous            | ✅/⚠️  | [Constituents and weights defined] |
| Reflective of knowledge| ✅/⚠️  | [Manager familiar with holdings]   |
| Investable             | ✅/⚠️  | [ETF proxy: [TICKER]]             |
| Owned by manager       | ✅/⚠️  | [Manager acceptance: pending]      |

━━━ HISTORICAL VALIDATION ━━━
| Metric                     | Value    | Acceptable Range |
|----------------------------|----------|-----------------|
| Portfolio-Benchmark Corr   | 0.XX     | >0.90           |
| Annualized Tracking Error  | XX.X%    | X-X%            |
| Information Ratio          | X.XX     | >0.00           |
| Avg Monthly Excess Return  | X.XX%    | —               |
| Active Share               | XX.X%    | 20-80%          |

━━━ BENCHMARK SPECIFICATION (IPS LANGUAGE) ━━━
"The portfolio shall be measured against [benchmark name/blend], rebalanced [frequency],
calculated on a total-return basis in [currency]. The benchmark is composed of:
[list components and weights]. The benchmark was selected because [rationale].
Performance will be evaluated over rolling [3/5]-year periods."

━━━ TRIPLE-THREAT LENS ━━━
🏦 **Banker:** [Is this benchmark appropriate for marketing materials? Would institutional investors accept it? How does it compare to what peers use? Is there a standard benchmark the consultant community expects?]
📊 **Accountant:** [Is the benchmark return independently verifiable? Are the constituent weights transparent? Does the rebalancing methodology introduce return drag vs. the portfolio? Tax lot implications of benchmark replication?]
💰 **Wealth Manager:** [Can the client understand this benchmark? Is there an investable ETF proxy for the "just buy the benchmark" conversation? What's the expected tracking error budget — will the client tolerate periods of underperformance?]

━━━ SOURCES ━━━
[List all index providers, data sources, methodology documents referenced]

Quality Gates

• Portfolio mandate fully characterized (asset class, geography, cap, style, constraints)
• At least 3 candidate benchmarks evaluated with quantitative fit metrics
• CFA benchmark properties checklist completed — all 7 properties assessed
• Overlap, correlation, and tracking error calculated against portfolio history
• If blended: weights sum to 100%, each component ≥5%, maximum 5 components
• Benchmark returns independently reproducible from constituent data
• IPS-ready specification language provided
• Investable proxy identified (ETF or replicable strategy)
• Gaming/appropriateness check performed — benchmark isn't cherry-picked
• Historical validation shows reasonable tracking error and correlation
• Triple-threat lens references specific fit metrics

Professional Standards

What separates A from B:

• A-grade: Multiple candidates evaluated with quantitative fit metrics. CFA properties checklist completed. Blended benchmark has documented rationale for each component and weight. Historical validation shows the benchmark is a fair measuring stick. Gaming detection performed. IPS-ready specification provided.
• B-grade: Single index selected based on asset class match only. No overlap or correlation analysis. No documentation of why alternatives were rejected. No CFA properties assessment.

Common pitfalls:

• Choosing a benchmark after seeing results (backfitting the benchmark to flatter performance)
• Using an investable ETF as benchmark when it has significant tracking differences from the underlying index
• Blending too many indices — a 7-component benchmark is impossible to explain to clients
• Ignoring currency effects in international benchmarks
• Using peer group benchmarks without addressing survivorship bias
• Selecting a broad market index for a concentrated sector fund
• Not rebalancing the blended benchmark — drift causes the benchmark to deviate from the original intent

See Also

• backtest-engine — test strategies against the constructed benchmark
• concentration-risk-report — assess active risk vs benchmark
• multi-currency-portfolio — currency considerations for international benchmarks