From Financial State to Financial Trajectory: Detecting Deterioration Before It Shows

When a company fails, the warning signs usually look obvious in hindsight. Margins narrowed. Working capital tightened. Payments to suppliers slowed. Debt grew faster than revenue. Looking back, the path to distress is clear.

Looking forward, it rarely is. At any given moment, each of those indicators may still be within acceptable ranges. Credit analysts see a company that is, by every standard measure, in reasonable condition, until suddenly it isn't.

The problem isn't that the indicators are wrong. It's that they describe where a company is, not where it is going. Financial analysis has been built to measure state. Deterioration is a trajectory.

This article describes a research direction at Zayon, which we call STEPA, built on a simple hypothesis: an adequate representation of a company's financial trajectory can identify latent states of deterioration before they become evident in traditional indicators.

The limits of state-based analysis

Traditional financial analysis evaluates snapshots. A balance sheet at a date. An income statement for a period. Ratios computed from both: leverage, liquidity, coverage, profitability. Credit models built on these inputs inherit the same perspective, even when they include a few lagged values.

This approach has three structural limitations.

• Snapshots hide direction. Two companies with the same leverage ratio today may be in opposite situations: one steadily deleveraging, the other accumulating debt quarter after quarter. A state-based view treats them as equivalent.

• Thresholds trigger late. Early warning systems built on state typically fire when a ratio crosses a threshold. By the time a liquidity ratio breaches its limit, the deterioration that caused it has often been under way for many quarters.

• Ratios move together, and the pattern matters. Deterioration rarely shows up in one indicator. It appears as a pattern of small, coordinated changes across many: a slight decline in margins, a modest lengthening of receivables, a gradual shift in the funding mix. Each change alone is within normal variation. Together, they describe a direction.

From state to trajectory

A trajectory approach changes the unit of analysis. Instead of asking "what is this company's condition now?", it asks "what path is this company on, and what does that path resemble?"

That requires representing a company not as a point but as a movement through a space of financial characteristics over time. Two companies at the same point today can be on very different trajectories. Two companies at different points can be on the same one.

The research question is whether trajectories can be represented in a way that separates healthy evolution from early deterioration, and whether that separation appears earlier than it does in state-based indicators.

The research approach

STEPA explores this question through learned representations of financial trajectories. At a high level, the approach has four elements.

• Sequences, not snapshots. The input is a sequence of a company's financial characteristics over time, not a single period. The model learns from how characteristics evolve together.

• Contrasting healthy and deteriorating paths. The representation is learned so that trajectories of companies that remained healthy and trajectories of companies that later deteriorated occupy distinguishable regions, even at stages where their individual ratios still look similar.

• Macroeconomic conditioning. A company's trajectory can't be interpreted without its environment. A margin decline during a sector-wide downturn means something different from the same decline in a stable economy. The representation is conditioned on macroeconomic context, so that it distinguishes company-specific deterioration from shared cyclical movements.

• Stable, informative representations. The learning process is regularized to produce representations that are robust and that preserve meaningful variation, rather than collapsing into a few uninformative patterns.

• The specific architecture, features and training methods are part of Zayon's proprietary research and are not described here.

What it could change

If the hypothesis holds, the practical implications for financial institutions are significant.

• Earlier intervention windows. The value of early warning is measured in time: the earlier deterioration is detected, the more options an institution has. Reviewing a limit, requesting additional collateral or restructuring an obligation is far more effective two quarters before distress than two weeks before.

• Better-targeted monitoring. Institutions monitoring large portfolios can't review every company closely. Trajectory signals could help prioritize analyst attention toward companies whose paths are changing, rather than those that merely look weak today.

• Distinguishing cyclical from structural. Conditioning on macroeconomic context could help separate companies affected by a temporary, shared downturn from those deteriorating for company-specific reasons, which call for very different responses.

• Supervisory applications. The same approach can, in principle, be applied at a higher level, to monitor trajectories of institutions or segments rather than individual companies, supporting supervisory prioritization.

What it doesn't change

A trajectory signal, however early, is still a signal. It doesn't decide anything.

In a governed decision system, a STEPA-type signal would be one piece of evidence, with its uncertainty explicitly attached, feeding a decision governed by the institution's policy. The policy determines what an early warning means: whether it triggers a review, an adjustment, an escalation to an analyst, or nothing at all. An accountable person or committee decides on interventions with real consequences for the company.

That separation matters especially for early signals. The earlier a signal fires, the more uncertain it tends to be. Acting on early, uncertain signals without a policy that accounts for uncertainty would trade late detection for premature, poorly justified intervention. Neither serves the institution or the companies it lends to.

Research, not a promise

STEPA is research. It is grounded in a clear hypothesis and a defined method, and it is being developed with the rigor that implies. We present it as a direction, not a finished product with proven results.

Research results in this field require careful validation: across economic cycles, sectors and company sizes, with attention to how early signals perform in practice and how often they are wrong. We will share results as they are validated, with their limitations stated as clearly as their strengths.

What we are confident about is the question itself. Financial analysis has spent decades refining how to measure where a company is. The next step is learning to see where it is going.

From financial state analysis to financial trajectory intelligence.

Interested in Zayon's research on financial trajectories? Talk to our team →