Causal identification: analyst primer
Scope of identified effects by module, vocabulary crosswalk, and suggested IC language — without overstating inference strength.
Identification map: causal vs. structural vs. heuristic
- CPA (Portfolio) — Genuine causal decomposition. Most valuable for regime robustness.
- M&A — Genuine: DAG + do-calculus (E[Efficiency | do(merge)]).
- Product Fit — MCDA + causal ROI adjustment (confounding). Core fit scoring is not causal.
- Investment (CVC) — Composite heuristic with decision structure; not formal causal inference.
Interventional objective
Descriptive stacks stop at "what correlated with what." Where identified, we target "what shifts under an assignment or policy,"i.e. quantities aligned with do-operator semantics rather than vendor case-study selection.
Terminology crosswalk
| Formal term | Practitioner gloss |
|---|---|
| Confounding / confounders | Other factors that mix with the thing you care about. Example: "Big CUs get better results" — size might be driving results, not just the vendor. |
| Heterogeneous treatment effect (HTE) | The effect varies by type of institution. "Works well for large CUs, less for small ones." |
| Adjusted ROI / ATT | We take vendor claims and adjust them for your situation (size, digital maturity, data quality) instead of assuming you match their best-case study. |
| Causal credibility | How much we trust vendor case studies. Low credibility = likely selection bias, cherry-picked success stories. |
| Counterfactual | "What if we did something different?" — e.g. "What if we don't adopt?" or "What if rates go up?" |
| Stress test / scenario | Simulate a shock (e.g., Fed raises rates) and see what happens to your portfolio. |
| Causal diversification score (CDS) | How much your portfolio is diversified across different drivers, not just across sectors. High CDS = less hidden concentration in the same risk. |
| Σ_obs vs Σ_do | Observed links vs links after stripping shared macro drivers. Shows how much co-movement is "real" vs driven by common shocks. |
Vendor Adoption (Adjusted ROI)
What we answer: "Will this vendor work for us?"
Vendors show their best customers. We compare your CU (size, digital adoption, data quality) to those cases and adjust the expected benefit up or down.
Suggested report language
"We adjusted vendor-reported outcomes for your institution's size, digital adoption, and data quality. The adjusted ROI reflects our best estimate for your situation, not for their ideal customer profile."
Investment Portfolio (CDS)
What we answer: "How will our portfolio behave if things change?"
We model how your assets respond to rates, volatility, inflation. CDS measures how diversified you are across different risk drivers. Higher CDS = less hidden concentration in one risk.
CDS in one line
"How much your diversification is spread across different drivers, not just sectors."
Module selection (workflow)
| Use case | Primary deliverable |
|---|---|
| Technology / vendor diligence | Heterogeneity-adjusted ROI lattice vs. charter baseline |
| Balance-sheet / sleeve review | Σ_obs / Σ_do decomposition and CDS readout |
| Monetary tightening or volatility regime | Scenario lattice on macro state variables |
| Status-quo vs. adoption | Counterfactual no-adopt path vs. interventional adopt path (where modeled) |
Full guide: docs/CAUSAL_ANALYSIS_LAYMAN_GUIDE.md