Covenant tests (ratios.covenants)¶

Answers the question "does the company respect its covenants?". Tests a list of financial covenants, each on the numerator and the denominator that the loan agreement defines, against a threshold at a test date, and returns for each covenant an explicit outcome (passed, breached or not_tested with a code and the reason), the tested value and the headroom.

The method does not compute the standard ratios of the other groups: a covenant is tested on the agreement's own definitions (for example "consolidated EBITDA for the 12 months ending on the test date, before non-recurring items"), which often differ from the reported figures. The analyst computes the amounts under those definitions and gives them with the words of the agreement; Fairmark echoes both in the audit trail.

Fairmark never decides what the loan agreement decides. Where agreements differ, the clause is an explicit input with no default: the outcome when the denominator is zero or negative (if_denominator_not_positive) and the rounding of the tested value (rounding). Every outcome therefore traces back to a clause of the agreement.

Covenant tests and analytical ratios¶

A covenant test can give a different number from the analytical ratio of the same name in ratios.debt_service, and both are correct for their purpose. The analytical ratio uses the reported aggregates under the conventions the analyst chooses (balances, EBITDA and interest definitions). The covenant uses the definitions written in the loan agreement, which commonly adjust EBITDA (non-recurring items, pro forma effect of acquisitions and disposals), include or exclude items of debt (leases, shareholder loans), cap the cash that may be netted, and fix the test period (often the 12 months ending on the test date). The lender reads compliance only from the agreement's definitions, so the covenant test never substitutes the analytical ratio for them.

Formula¶

The headroom is positive on the compliant side of the threshold and negative on the breach side, whatever the direction of the covenant. At the threshold it is zero: an inclusive operator (at_most, at_least) passes and a strict one (less_than, greater_than) breaches.

Precision. The outcome is decided exactly. Without rounding, it is the sign of numerator − threshold × denominator, computed exactly in integer arithmetic (the denominator is positive at that point), so a tested value that rounds to the threshold at 28 digits is still classified on its exact value; value and headroom are shown to the 28 significant digits of the decimal type, not rounded to the decimals of the agreement. With rounding, the tested value is rounded as the agreement states, computed from the exact quotient (no intermediate rounding), and the outcome is the exact comparison of that rounded value with the threshold; value is the rounded value and value_unrounded the value before rounding. When the displayed headroom rounds to 0 at 28 significant digits (or shows the other sign) while the exact comparison decides otherwise, a convention and an explain note say so: "displayed headroom rounds to 0 at 28 significant digits; the exact comparison decides breached".

Zero or negative denominator. The tested value is not computed: value and headroom are null and the outcome is the one stated by if_denominator_not_positive (breached, passed or not_tested), and outcome_basis is if_denominator_not_positive, so an agent can tell an outcome set by the agreement's rule from one decided on a tested value (tested_value). With not_tested the code is denominator_not_positive and the reason says whether the denominator is zero or negative.

Note: headroom_relative is headroom / threshold. For a maximum covenant with a positive tested value it is also the fraction by which the denominator can fall, the numerator unchanged, before the covenant is breached. For a minimum covenant the fraction by which the numerator can fall, the denominator unchanged, is headroom / value (for example a minimum interest cover of 4 tested at 5: headroom 1, the numerator can fall by 1 / 5 = 20%, while headroom / threshold is 25%). Fairmark does not return headroom / value as a separate field; it follows from headroom and value.

Example¶

from decimal import Decimal

import fairmark as fm

ltm = "for the 12 months ending on the test date"
r = fm.ratios.covenants(
    covenants=[
        {
            "name": "net leverage",
            "test_date": "2026-12-31",
            "numerator": {"definition": "consolidated net financial debt", "amount": 1900},
            "denominator": {"definition": f"consolidated EBITDA {ltm}", "amount": 600},
            "operator": "at_most",
            "threshold": "3.5",
            "if_denominator_not_positive": "breached",
        },
        {
            "name": "interest cover",
            "test_date": "2026-12-31",
            "numerator": {"definition": f"consolidated EBITDA {ltm}", "amount": 600},
            "denominator": {"definition": f"net finance charges {ltm}", "amount": 175},
            "operator": "at_least",
            "threshold": "4",
            "if_denominator_not_positive": "passed",
        },
        {
            "name": "debt service cover",
            "test_date": "2026-12-31",
            "numerator": {"definition": f"cash flow available for debt service {ltm}", "amount": 700},
            "denominator": {"definition": f"debt service {ltm}", "amount": 0},
            "operator": "at_least",
            "threshold": "1.2",
            "if_denominator_not_positive": "not_tested",
        },
        {
            "name": "senior leverage",
            "test_date": "2026-12-31",
            "numerator": {"definition": "senior net financial debt", "amount": 3504},
            "denominator": {"definition": f"consolidated EBITDA {ltm}", "amount": 1000},
            "operator": "at_most",
            "threshold": "3.50",
            "if_denominator_not_positive": "breached",
            "rounding": {"decimal_places": 2, "mode": "half_up"},
        },
    ]
)
leverage, cover, dscr, senior = r.covenants
assert leverage["outcome"] == "passed"  # 1900 / 600 = 3.1666... <= 3.5
assert abs(leverage["headroom"] - Decimal("0.3333333333")) < Decimal("1e-10")
assert cover["outcome"] == "breached"  # 600 / 175 = 3.4285... < 4
assert cover["headroom"] < 0  # negative: on the breach side
assert dscr["outcome"] == "not_tested"  # zero debt service, as the agreement states
assert dscr["not_tested_code"] == "denominator_not_positive"
assert dscr["outcome_basis"] == "if_denominator_not_positive"
assert leverage["outcome_basis"] == "tested_value"
assert dscr["not_tested_reason"].startswith("denominator is zero")
# 3.504 rounds to 3.50 under the agreement, which passes at most 3.50
assert senior["outcome"] == "passed"
assert senior["value_unrounded"] == Decimal("3.504")
assert r.summary == {"passed": 2, "breached": 1, "not_tested": 1}

Inputs¶

Outputs¶

Conventions (explicit, reported in explain.conventions)¶

Covenants are numbered from 1 in the audit trail. Each covenant adds its own test rule ("covenant 1, net leverage (2026-12-31): passes when the tested value ≤ 3.5 (maximum, inclusive)"), its rule for a denominator that is not positive and its rounding rule, both marked as coming from the input, a verdict step ("breached (600 / 175 at_least 4 is false, decided exactly)"), the unrounded and rounded value when rounding is given and, when the covenant is not tested, the reason; the three counts of the summary are steps too. A negative numerator is compared as computed: a net cash position gives a negative leverage, which passes a maximum; a negative EBITDA gives a negative cover, which breaches a minimum.

Not covered¶

• Covenants on a single amount (minimum net worth, maximum capital expenditure) have no denominator and are not tested by this method; see the roadmap.
• Equity cures are not modelled in version 1.
• Step-down schedules are expressed as one covenant per test date, each with its own threshold.
• The relative headroom on a negative threshold is not defined.

Errors¶

Accuracy¶

The outcome is exact: without rounding it compares numerator and threshold × denominator as integers; with rounding the rounded value is computed exactly from the quotient of the two amounts and compared with the threshold. The tested value is one decimal division, the headroom one subtraction and the relative headroom one division, each to the 28 significant digits of the decimal type.

Verified against¶

• Golden cases in tests/golden/ratios.covenants/, whose expected values come from scripts/reference_ratios.py (Python fractions for the outcome and the rounding, decimal at 50 digits for the printed values): a covenant set with one breach and the headroom of the others, a zero and a negative denominator under each value of if_denominator_not_positive, 3.504 tested at most 3.50 with and without rounding to 2 decimals, half_up against half_even on a half, both operators of each direction at the threshold, negative numerators and non-positive thresholds; a constructed case for overflow.
• Unit tests on values that round to the threshold at 28 digits, for example 3.0000000000000000000000000001 / 3 tested at most 1, which is breached, and on a quotient whose 28-digit rounding would round the wrong way at 2 decimals.