Backtest — Calibration vs Rated Mauritian Firms

Backtest · Calibration vs Rated Mauritian Firms

Runs the same engine as the platform against firms with a real CRAF / GCR public rating

Read this first. CRAF and GCR publish national-scale ratings (e.g. CARE MAU A+, AAA(MU)) that are relative within Mauritius and sit several notches above their global-scale equivalent. This engine outputs global-scale bands, capped at the Mauritius sovereign (Baa3). So a strong Mauritian corporate correctly lands at ~BBB- here even when its national rating is AA/AAA. The valid comparison is therefore rank-ordering, not the letter itself — we test whether the engine ranks these firms in the same order the agencies do. (GCR’s own data confirms the gap: Forty Two Point Two is AAA(MU) but only BBB- global.)

How to use

Firms with financials

of 0 in roster

Rank agreement (Spearman ρ)

—

engine score vs published rank

Within sovereign cap

—

firms hitting Baa3 ceiling

Self-test

—

demo corp engine band

The engine needs financials, which are not published in the agencies’ rationale PDFs. Pull them from each firm’s annual report / SEM filing and enter them per row (click Edit financials). Entries are saved in your browser (localStorage). Qualitative inputs are held neutral so this is a financials-driven calibration. The Self-test (Demo Corp) row is pre-filled so you can see the mechanism working immediately. As you add real firms, the Spearman ρ tells you whether the engine ranks them like CRAF/GCR do (ρ → 1 = perfect rank agreement). Rows with an unconfirmed published symbol are shown but excluded from ρ.

Roster

Sources are linked per row. National-scale symbols verified against a real CRAF rationale or GCR announcement; unconfirmed notches are flagged — do not treat them as exact.

Company	Sector	Agency	Published (national)	Nat. rank	Engine band (global)	Engine score	Actions

Reading the result

If ρ is high (>0.7): the engine ranks Mauritian credits the way the licensed agencies do — strong evidence the scorecard is well-calibrated for orientation, even though it deliberately reports a lower (global) letter. If ρ is low or negative: investigate which firm the engine mis-ranks and why (often a sector the generic scorecard handles poorly — use the hospitality template, or build a sugar/agro template next).

ρ uses average-rank tie correction (Pearson on tied ranks) — many firms share a national notch (e.g. five at "A"), which biases the simple Spearman formula. National rank vs engine composite; sign flipped so agreement → +1.

Current sample: 14 firms with audited financials, tie-corrected ρ ≈ 0.77. This is the strongest result so far, and the reason is range: adding genuine sub-investment-grade anchors (Mammouth BBB+, Chartreuse BBB, Cap Tamarin BBB-) finally gave the engine room to discriminate. It now ranks weak credits correctly — Cap Tamarin (loss-making property dev) lowest at B+, the sugar names BB/BB+, conglomerates and healthcare at the top — across a real AA→BBB- spread instead of a compressed A-cluster.

Honest caveats remain: (1) the engine still disagrees with CARE on a few names for defensible reasons (NMH tourism-cyclical, Medine asset-rich/cash-poor) — not tuned away; (2) several BBB-band figures come from CRAF summary tables that omit current/total assets, so their liquidity ratios are under-scored (flagged per firm); (3) 14 firms is still small. Treat ρ ≈ 0.77 as strong directional validation, not statistical proof. The lowest clean CARE corporate (Le Collège de l'Ouest, BB+) is a project-stage SPV with no issuer financials, so it still cannot be scored.