Publications

Research

The BYLAS™ Engine rests on two preregistered papers, co-authored with Hitotsubashi University ICS and published on SSRN. Everything below is reproducible from frozen, hash-verified analysis files.

What the research establishes

Behavioral decision parameters carry information about a person that observable attributes (age, role, experience, education) do not explain. Predicted in the first paper, confirmed in the field in the second.
Loss aversion is not fixed. Under a crisis frame it rises by about a quarter, and about 90% of people move in the same direction. A measurement taken once, in calm conditions, misses nearly half of the people who become most loss-averse under stress.
The measurement is preregistered, the instrument is frozen under a cryptographic hash before data collection, and every published number reproduces from the archived files.

Working PaperSSRNNew
Loss Aversion Is a State, Not Just a Trait
Preregistered Model-Free Evidence from the Field
Yasuhiro Kawakami (Pensata, Inc.) · Ryosuke Fujitani (Hitotsubashi University ICS)
Posted September 30, 2026 · DOI: 10.2139/ssrn.7544500 · Preregistered on OSF (n2mw9)
Why it matters for lenders
Defaults happen under stress, not in calm conditions. This study shows that a risk profile measured once, in the normal state, misses nearly half of the people who become most loss-averse when conditions turn. The BYLAS Engine measures both states in one short session.
Key findings
- Preregistered field experiment with 143 employees of a major Japanese financial group, each measured under a normal frame and then a crisis frame in a single session.
- Loss aversion rose by about 25% under the crisis frame; roughly 90% of participants moved in the same direction.
- Of the participants most loss-averse under stress, 46% could not be identified from their normal-state measurement alone.
- The shift is visible in raw choices, without relying on any estimation model: when offered strictly better odds under the crisis frame, participation fell instead of rising, while a control pair did not move.
- Every reported figure reproduces from an analysis archive frozen under SHA-256 hashes.
Companion to “Beyond FICO” (SSRN, August 2026): this paper provides the field evidence for the two-state measurement framework introduced there.
Abstract
Is loss aversion a trait, or a state? We report a preregistered test: Japanese financial-group employees completed a two-state task (Kawakami and Fujitani, 2026) in one sitting, under a normal and then a crisis frame. Because selected stress items offer strictly better odds, a monotone choice model with unchanged preferences and reference points and no position effects predicts nondecreasing expected participation. It fell, significantly; a matched control pair did not. Loss aversion rises under stress. Three of four registered hypotheses are supported; the attribute bound is falsified for probability weighting alone. Normal-state rank predicts stress-state rank only in part, so single-state measurement misses a state component that the attribute set does not reach.
Keywords: loss aversion, prospect theory, state-dependent preferences, risk preference stability, preregistration, field experiment · JEL: C93, D81, D91, G41
Read on SSRN
Working PaperSSRN
Beyond FICO
Behavioral Calibration of SME Cash Flows and the Future of Revenue-based Financing: Toward Financial Inclusion for Visionaries, Small Businesses, and the Vulnerable
Yasuhiro Kawakami (Pensata, Inc.) · Ryosuke Fujitani (Hitotsubashi University ICS)
Posted August 11, 2026 · DOI: 10.2139/ssrn.7235778 · Preregistered on OSF (n2mw9)
Why it matters for lenders
Credit files score history. This paper lays out how to score the person: the behavioral signals that the BYLAS Engine measures, why they are features to price rather than risks to screen out, and how a thin-file applicant’s own projections can be read through them.
Key findings
- Reframes behavioral biases as measurable features of the borrower rather than deviations to penalize.
- Derives a closed-form map from an entrepreneur’s own cash-flow projections to objective firm fundamentals, so a business plan can be corrected for the profile of the person who wrote it.
- Predicts that applicants who look identical on observable attributes separate into distinct behavioral profiles, and that attributes explain only a small share of that variation. This prediction was confirmed in the field in the companion paper.
- Shows how the measured profile can be used to match a borrower to a product structure, including revenue-based financing variants and micro-credit lines, rather than to ration access.
- Pre-registers the two-state measurement design and four falsifiable hypotheses before any data were collected.
Field evidence for this framework is reported in “Loss Aversion Is a State, Not Just a Trait” (SSRN, September 2026).
Abstract
The paper uses the name “BYLAS Score” for the measurement that is now delivered as the BYLAS Engine.
Traditional credit scoring, notably FICO, penalizes the volatile cash flows typical of entrepreneurs and small and medium-sized enterprises (SMEs). We propose the BYLAS Score, which treats prospect-theoretic biases not as risks but as stable, measurable features of the borrower. Embedding risk sensitivity, loss aversion, and probability weighting in a continuous-time cash-flow model yields a closed-form map from biased projections to objective firm fundamentals (μ, σ), which in the baseline regime reduces to a single γ-dependent adjustment to the stated drift. The instrument—which measures each subject twice within a single sitting, under a normal frame and an induced-stress frame—the identification strategy, and the hypotheses are pre-specified in advance of data collection. The central prediction is that subjects near-identical on observable attributes nonetheless separate into distinct behavioral archetypes, with those attributes explaining only a small share of the behavioral variance. We draw out the implications for credit-thin and low-FICO populations, proposing tailored revenue-based financing, micro-credit, and ESG guardrails.
Keywords: behavioral finance, prospect theory, credit scoring, financial inclusion, SME lending, revenue-based financing, BYLAS Score · JEL: G21, G41, D81, O16
Read on SSRN

Want to see this run on your decline flow?

Shadow validation runs inside your live origination flow with no change to your credit decision. We take the risk first.

Request a Meeting

Research

Loss Aversion Is a State, Not Just a Trait

Beyond FICO

Want to see this run on your decline flow?