How to Find Companies at Risk of Default With Data

Combining filing, balance-sheet, and price signals to predict bankruptcy data before the market prices in the risk.

Most defaults do not arrive without warning. The signals accumulate across quarterly filings, debt-market pricing, and regulatory disclosures for months — sometimes years — before a company misses a payment or files for protection. Building a systematic approach to identify that accumulation is what separates opportunistic discovery from being caught flat-footed. The data required to predict bankruptcy data already exists in public sources; the work is constructing a multi-signal framework that combines it correctly and pulls it without look-ahead bias.

A well-labeled historical record of outcomes is the starting point. Alphanume's corporate default events dataset provides dated records of payment defaults, covenant violations, and bankruptcy filings — the ground truth needed to backtest any watchlist methodology and validate that early-warning signals are actually early.

Fundamental distress signals

Balance-sheet deterioration tends to be the earliest measurable signal. The four most actionable ratios to track at a quarterly frequency:

• Interest coverage ratio below 1×. Operating income insufficient to cover interest expense means the company is technically consuming capital to service debt. A sustained reading below 1× is a hard trigger.
• Negative or rapidly shrinking cash runway. Free cash flow burn divided into cash and equivalents gives quarters of liquidity remaining. When that figure drops below three to four quarters, the refinancing window narrows sharply.
• Rising net debt / EBITDA. Leverage expansion in the same period that EBITDA is contracting compresses the debt-service cushion from both directions. Track the trend, not just the level.
• Negative working capital. Current liabilities exceeding current assets can signal an inability to meet near-term obligations without rollover credit.

The Altman Z-score aggregates several of these dimensions into a single composite. The Altman Z-score uses five weighted ratios — working capital, retained earnings, EBIT, market cap, and sales, each scaled by total assets — to produce a score below which distress probability rises materially. The original public-company model treated scores below 1.81 as the distress zone. The score is useful as a screener but not as a standalone predictor; it should feed a broader model alongside the signals below.

Disclosure signals from regulatory filings

SEC filings contain explicit distress language that most equity screens ignore entirely. These are point-in-time signals — they carry the actual filing date, which matters for avoiding look-ahead bias.

• Going-concern opinion. When an auditor adds a going-concern explanatory paragraph to a 10-K or 10-Q, it is a formal judgment that substantial doubt exists about the company's ability to continue as a going concern for twelve months. This is a high-specificity signal.
• Covenant-breach disclosures. Companies disclose covenant violations in 10-Q and 10-K notes, typically under the long-term debt footnote. Waivers and amendments are also disclosed — a pattern of repeated waivers is more informative than a single breach.
• 8-K defaults. Item 1.01 and 2.04 8-K filings explicitly disclose material agreements and triggering events — including missed payments and acceleration notices. These are often filed after the fact but before equity markets fully react.
• Auditor changes. A sudden change in auditor, particularly when accompanied by a disclosure of disagreements under Item 4.02, warrants scrutiny.
• Late filings and NT forms. A Form NT 10-K or NT 10-Q signals that the company cannot file on time. Late reporting correlates strongly with accounting issues and operational distress in the period leading up to default.

Market-price signals

Capital-market prices incorporate information before filings do. Three channels to monitor:

Bond pricing and CDS spreads. When a company's bonds trade at a significant discount to par — 70 cents or below is a common threshold for distressed categorization — the debt market is pricing in meaningful default probability. CDS spreads on the same issuer, where available, provide a continuous read on the market's probability estimate. Bond prices are often the fastest-moving distress signal.

Equity price decline and volatility. A sustained drawdown combined with rising implied volatility reflects option-market participants hedging or speculating on a binary outcome. Equity is junior to all debt in the capital structure, so deep equity declines often lag bond deterioration but confirm it.

Hard-to-borrow status. When a stock becomes difficult to borrow and short interest rises sharply, it indicates that informed market participants are positioning for further downside. Securities lending data is a useful confirming signal rather than a primary one.

Capital-structure overhang

The structure of liabilities matters as much as their magnitude. A company with $500 million in senior secured debt maturing in eight months is more immediately at risk than one with $2 billion in investment-grade bonds maturing over a decade, even if the second company looks worse on leverage ratios. Key structural indicators:

• Near-term debt maturity concentration. The maturity schedule from debt footnotes or the schedule of long-term debt shows exactly when principal is due. A wall of maturities within 12 to 24 months, against thin cash balances and closed capital markets access, is a default precursor.
• Secured vs. unsecured mix. Heavy reliance on secured borrowing — revolving credit, term loans — can limit a company's ability to raise additional secured debt when it needs liquidity.
• PIK toggles and deferred interest. Payment-in-kind provisions allow interest to accrue rather than be paid in cash. Companies electing PIK are conserving cash under duress; the compounding balance makes eventual resolution harder.

Building a ranked watchlist without look-ahead bias

Combining these signals into a usable watchlist requires strict data discipline. The central problem is that financial data is frequently restated, and vendors backfill corrected figures. Using restated data to predict defaults on historical dates overstates model performance — the data as available on the filing date must be used, not the data as it exists in the database today. Point-in-time snapshots of fundamentals, using the original-filing dates and original-reported figures, are the correct source for any training or backtest exercise.

A practical signal-combination approach:

1. Score each fundamental dimension. Assign ordinal scores to interest coverage, leverage trend, cash runway, and Altman Z. This normalizes across different units and scales.
2. Flag filing-based triggers. Going-concern opinions, NT filings, and 8-K default items are binary flags that automatically escalate any name on the watchlist.
3. Layer in market signals. Bond prices below par and equity drawdown add recency — market signals update continuously, while filings update quarterly.
4. Apply a maturity-proximity multiplier. Names with large maturities in the next 12 months receive additional weight regardless of other scores.

To build a distress screener that is statistically valid, the resulting ranked list should be validated against a labeled default event history: what fraction of names in the top decile of the score actually defaulted within 12 months? That precision metric, computed out-of-sample, is what separates a useful watchlist from noise.

Managing false positives

The unavoidable feature of any distress screen is that most names on it will not default. Companies restructure out of court, raise dilutive equity, sell assets, or simply survive on minimal cash flow for longer than models predict. A framework that treats every distressed signal as a default prediction will generate far more false positives than actionable situations.

Two adjustments reduce false-positive costs:

• Require signal convergence. A name should appear in multiple signal dimensions — not just weak fundamentals, but also deteriorating bond prices and a recent filing trigger — before it is escalated to high-conviction status.
• Track trend, not level. A company with an interest coverage ratio of 0.8× that has been stable for six quarters is in a different situation from one that was 2.5× a year ago and has deteriorated rapidly. Rate of change is often more predictive than the absolute reading.

Historical default events, mapped back through the signal data to 12 and 24 months prior, show what early-warning combinations actually preceded default versus which combinations resolved without one. That labeling exercise — connecting outcomes to precursor signals — is the core of converting a watchlist into a model that can be evaluated and improved over time.