Seven financial ratios predict SME bankruptcies up to three years in advance
There is a peculiar moment in any sector when the evidence that would solve a problem has been available for decades, but nobody had organised it in the right way. That is, in essence, what a study just published in the Global Business and Economics Review has documented: that the insolvency of small and medium-sized enterprises in Europe can be anticipated up to three years in advance using only seven standard accounting indicators — the very same ones that any accountant already calculates and that banks have been receiving for decades without knowing exactly what to do with them in combination.
The study, authored by Sónia Silva, analysed data from more than 24,500 European companies over eight years. The resulting model achieves an overall accuracy of approximately 82% and manages to correctly identify more than 70% of insolvencies three years before they occur, when applied to data with known outcomes. It is not a machine learning algorithm trained on millions of variables. It is a multivariate prediction model built on seven ratios: the cash ratio, the interest coverage contribution, the solvency ratio, short-term financing, leverage, the debt-to-assets ratio, and return on assets. Seven numbers that already exist in balance sheets and that, taken together, describe with sufficient fidelity the liquidity, the debt burden, the financial resilience, and the profitability of a company.
The question that this finding leaves on the table is not a technical one. It is structural: if the data were already there and the models work, what prevented this from happening sooner?
The gap that academia kept open for too long
The prediction of corporate bankruptcies has a long academic history. The classic models, conceived in the 1960s and 1970s, were designed for large, publicly listed companies with access to market valuations, stock market capitalisation data, and financial structures that were transparent enough to feed robust statistical models. SMEs were left outside that framework, not out of negligence, but because they represented a category that traditional corporate finance treated as too opaque, too heterogeneous, and in many cases too small to justify the analytical effort.
The problem with that logic is that SMEs are not a minor category. They represent the majority of companies in OECD economies and approximately two thirds of employment in those countries. The risk of insolvency among SMEs is not a microeconomic problem manageable at an individual scale: it is a variable with direct consequences for the banking system, the labour market, and the fiscal stability of governments that operate credit guarantee programmes or employment subsidies.
What Silva's work does is close that gap with a dataset large enough for the model to be statistically robust and sufficiently concentrated on accessible ratios for it to be replicable without extraordinary infrastructure. The most revealing finding is not that the model works with 82% accuracy: it is that this level of accuracy is achieved three years before the insolvency event, a time horizon that completely changes the logic of intervention.
Three years is enough time to renegotiate credit conditions. It is enough time for a lender to adjust guarantees, modify covenants, or intensify the monitoring of a specific portfolio. It is enough time for an SME owner to make restructuring decisions before the situation becomes irreversible. What distinguishes this model from the early warning systems that already exist in many European banks is precisely that extended horizon combined with the parsimony of the instrument: seven ratios, not hundreds of variables.
What the seven ratios reveal about the anatomy of a bankruptcy
Looking at the model's seven indicators as a set, rather than as isolated variables, produces a more interesting diagnosis than any one of them in isolation. The selection is not arbitrary: each ratio captures a distinct dimension of risk, and together they construct a three-dimensional picture of the company.
The cash ratio and short-term financing describe immediate liquidity and the way in which the company manages its most urgent obligations. A company can be profitable on paper and still suffocate from a lack of available cash. That is not an infrequent paradox in SMEs: it is one of the most common bankruptcy mechanisms, especially in businesses with long collection cycles and suppliers that demand rapid payment.
The solvency ratio, leverage, and the debt-to-assets ratio capture the capital structure and the capacity to absorb losses without collapsing. A highly leveraged company can survive as long as cash flows remain stable, but its margin of tolerance in the face of a revenue decline is minimal. These three ratios, viewed together, describe how much oxygen the company has left before its debt ceases to be sustainable.
The interest coverage contribution adds an operational dimension: it measures whether the business generates sufficient contribution margin to cover its financial cost. A company that cannot cover its interest payments with its operating margin is consuming equity or additional credit to stay active, which is a signal of structural deterioration that can remain invisible for several quarters if one looks only at net profit.
Return on assets closes the model by measuring the efficiency with which the company converts its assets into results. A sustained decline in this indicator, combined with rising leverage and declining liquidity, produces the pattern that the model learns to recognise as a precursor to insolvency.
What is significant from a value proposition perspective is that none of these seven ratios requires information that is not already available in a company's basic financial statements. There is no need for access to market data, external valuations, or management projections. The model operates with what already exists, which has direct implications for who can adopt it and at what cost.
The bottleneck that the model cannot resolve on its own
The study itself points to a limitation that deserves separate attention: the model would improve with greater financial disclosure by SMEs, but the authors consider this to be "highly unlikely given the nature of the smaller companies." That sentence concentrates a tension that is not new, but which this finding puts back on the table with greater urgency than before.
SMEs have structural incentives to keep their financial data opaque. Part of that opacity is defensive: sharing detailed information with lenders or with the market can weaken an owner's negotiating position, expose competitive vulnerabilities, or simply generate administrative burdens that a small company lacks the capacity to manage. The result is a market where prediction instruments work best precisely in the cases where information is most abundant, which tend to be the companies that need it least.
This imbalance has direct consequences for lenders. Banks and microcredit institutions operating in segments of small SMEs — not the well-documented medium-sized ones, but the micro and small enterprises with simplified accounting — have access to only a fraction of the information that the model requires to operate at its documented level of accuracy. In those cases, the model may still be useful as a relative risk reference, but its predictive capacity degrades in proportion to the quality of the available data.
For public credit guarantee programmes, the challenge is different but equally concrete. Many of these programmes operate under political pressure to maximise access to credit, which in practice means financing companies with risk profiles that a private bank would reject. A model with the accuracy documented by Silva could be used to better discriminate between viable companies with temporary liquidity problems and companies with irreversible structural deterioration, which would improve the efficiency of public spending. But that requires the beneficiary companies to report with the level of detail that the model needs, and that requirement clashes directly with the logic of simplification that justifies the programmes in the first place.
The data point that European banks should already be calculating
The research arrives at a moment when the European macroeconomic context amplifies its relevance. Previous studies on the impact of the COVID-19 pandemic on European SMEs documented insolvency risk increases of around 21% during that period, measured as a function of declines in profitability, turnover, and working capital. The very same variables that Silva's model identifies as central predictors.
For banks with significant SME lending portfolios, the economic argument for adopting a monitoring framework based on these seven ratios is straightforward. The International Monetary Fund has documented that widespread SME deterioration can reduce the Tier 1 capital ratios of banking systems by up to 2 percentage points in the most exposed countries. That is not an abstract risk: it is a variable that European regulators have been tracking with growing attention since 2020 and that the risk management teams of any bank with a relevant SME portfolio should be quantifying on a permanent basis.
The practical adoption of the model in bank monitoring systems does not require large technological investments. It requires discipline in the collection of periodic financial statements from borrowers, standardisation in the calculation of the seven ratios, and a clear internal alert process for when a company crosses risk thresholds across multiple indicators simultaneously. That is more a process problem than a technology problem, which materially lowers the implementation barrier for medium-sized institutions that do not have the budget for proprietary machine learning models.
What makes Silva's model especially useful in that context is not only its accuracy but its interpretability. A seven-ratio model is auditable. A credit analyst can explain to a risk committee why a specific company triggered an alert: "the cash ratio fell 40% over two consecutive financial years while leverage rose 15 percentage points and return on assets turned negative." That is a diagnosis that generates action. A black-box model with 200 variables may have greater statistical accuracy, but it produces more difficult conversations at the levels where real credit decisions are made.
The signal that nobody was reading together
The most enduring contribution of this work is not the model itself. It is the demonstration that the information needed to anticipate an SME's bankruptcy was already available, that it resided in the balance sheets that banks receive periodically, and that what was missing was the analytical structure to read it in combination with sufficient advance notice.
That describes a pattern that appears frequently in markets where data exist but are fragmented or poorly interpreted: the solution does not arrive with new information, but with a reorganisation of existing information that makes visible something that was already there. In this case, the reorganisation is statistically documented, replicable, and sufficiently parsimonious that any financial institution with access to basic balance sheets can adopt it without extraordinary infrastructure.
SMEs represent the majority of the business fabric of advanced economies and a disproportionate share of unmanaged credit risk. A model that can anticipate more than 70% of their insolvencies with a three-year margin, using only seven standard ratios, is not an academic curiosity. It is a tool with concrete operational consequences for lenders, regulators, and business owners who prefer to intervene rather than manage a crisis. The limit of its usefulness does not lie in its accuracy: it lies in the quality and consistency of the data that SMEs themselves are willing to report, and that depends on incentives that the model alone cannot change.
