← All projects

QRS research · Regulatory capital · Explainable finance

FRTBTrace

Exact trade-level attribution of FRTB-IMA capital

Trade-level explanations of FRTB market-risk capital, one question at a time.

The usual way to explain a bank’s FRTB market-risk capital trade by trade is to compute one attribution vector, typically a gradient or a published allocation rule, and use it for everything: desk reports, what-if screening, trade rankings. FRTBTrace separates the five questions that vector is asked to answer and computes each one exactly through the implemented internal-models calculation, keeping track of the past observation date each part of the capital comes from.

Details

Active

arXiv 2609.32238 · two ICAIF 2026 workshop papers, accepted

How FRTBTrace attributes capitalStored books on five past dates and today feed the capital graph: expected shortfall and the stress ratio, non-modellable and default-risk charges, a history maximum and the comparison with the standardised approach, ending in capital K. After one forward pass with its active branches recorded, one reverse pass sends the allocation back to every (date, trade) position. The ledger adds up to K; most of it sits on past dates, outside a sensitivity to today's book. Illustrative numbers.Stored books: date × tradet−5t−4t−3t−2t−1todaypast dates 63%today 37%row and column sums add up to K = 84.0ES, stress ratioper stored dateNon-modellablestress chargesDefault risk99.9% quantileHistory maxlatest vs averageavgSA comparisonmin{·}, (·)₊IMAK84.0Typed questions (removal and hedge choice by exact revaluation)MarginalRemovalAccountingOriginHedge choiceadds up to Kneed not add up1 · Capital and branch states2 · One reverse pass3 · Ledger on (date, trade)
How FRTBTrace attributes capitalStored books feed the capital graph, top to bottom, ending in capital K; one reverse pass sends the allocation back to every (date, trade) position, and the ledger adds up to K. Illustrative numbers.Stored books: date × tradet−5t−4t−3t−2t−1todaypast dates 63%today 37%ES, stress ratioNon-modellableDefault riskHistory maxavgSA comparisonIMAK84.0MarginalRemovalHedge choiceAccountingOriginThe two ledgers add up to K; the others need not.1 · Forward2 · Reverse3 · Ledger
One attribution, step by step: evaluate capital once through the graph, record the active branch at each floor and maximum, carry the allocation back in one reverse pass, and read the ledger on (date, trade) positions. Capital K is evaluated once through the graph over stored books; one reverse pass carries the allocation back to every (date, trade) position, and the ledger adds up to K. Illustrative numbers; the benchmark’s main books have 96 trades and 60 stored dates.

FRTBTrace at a glance

The idea

Write the whole capital calculation as one graph over (observation date, trade) positions. A forward pass records which branch is active at every floor, maximum and minimum; one reverse pass then carries the allocation back to every position. The resulting ledger adds up to capital:

∑s,iAs,iK=K

Sensitivities to today’s book miss exactly the part on past dates. Removal effects and hedge choices are answered by exact revaluation, and ties are flagged rather than hidden.

Experiments

Synthetic benchmark · 60 books tested

Capital on past dates, median61%
Range, 14 smooth main books21–99%
Ledger adds up to K, max error5.6 × 10⁻¹⁶
Benchmark-defined questions answeredall · others ≤ 2 of five

Pre-registered: 42 small tie books and 18 main books of 96 trades, compared with six published allocators and a numerical Euler control.

After the benchmark: Shapley values over trades under two declared conventions, and every question answered on a synthetic book of 960 trades.

Bank portfolios Coming soon

Every book so far is synthetic. Real portfolios with real desk groupings, an independent implementation of capital and timing measurements come next.

Capital on past observation dates, per main bookHorizontal bars for the 18 main books in six portfolio types: on the 14 smooth books, 21% to 99% of capital sits on past observation dates, median 61%. Books at a tie have no reference share.Share of capital K on past observation datesOrdinaryOrdinary, draw 1: 72.1% of K = 105.39 on past dates72%Ordinary, draw 2: 56.8% of K = 127.56 on past dates57%Ordinary, draw 3: 49.0% of K = 76.42 on past dates49%NonlinearNonlinear, draw 1: 48.8% of K = 95.42 on past dates49%Nonlinear, draw 2: 57.7% of K = 94.27 on past dates58%Nonlinear, draw 3: 56.0% of K = 79.59 on past dates56%Tail-difficultTail-difficult, draw 1: 91.9% of K = 454.17 on past dates92%Tail-difficult, draw 2: 64.3% of K = 352.53 on past dates64%Tail-difficult, draw 3: 75.1% of K = 311.01 on past dates75%History-heavyHistory-heavy, draw 1: 98.6% of K = 67.88 on past dates99%History-heavy, draw 2: 95.3% of K = 84.37 on past dates95%History-heavy, draw 3: 94.0% of K = 107.54 on past dates94%Held-outHeld-out, draw 1: 47.5% of K = 97.29 on past dates47%Held-out, draw 2: 21.0% of K = 93.62 on past dates21%at a tie: no reference answerRegulatory near-tieat a tie: no reference answerat a tie: no reference answerat a tie: no reference answermedian 61%0%50%100%
Capital on past observation dates, per main bookHorizontal bars for the 18 main books in six portfolio types: on the 14 smooth books, 21% to 99% of capital sits on past observation dates, median 61%. Books at a tie have no reference share.Share of capital K on past observation datesOrdinaryOrdinary, draw 1: 72.1% of K = 105.39 on past dates72%Ordinary, draw 2: 56.8% of K = 127.56 on past dates57%Ordinary, draw 3: 49.0% of K = 76.42 on past dates49%NonlinearNonlinear, draw 1: 48.8% of K = 95.42 on past dates49%Nonlinear, draw 2: 57.7% of K = 94.27 on past dates58%Nonlinear, draw 3: 56.0% of K = 79.59 on past dates56%Tail-difficultTail-difficult, draw 1: 91.9% of K = 454.17 on past dates92%Tail-difficult, draw 2: 64.3% of K = 352.53 on past dates64%Tail-difficult, draw 3: 75.1% of K = 311.01 on past dates75%History-heavyHistory-heavy, draw 1: 98.6% of K = 67.88 on past dates99%History-heavy, draw 2: 95.3% of K = 84.37 on past dates95%History-heavy, draw 3: 94.0% of K = 107.54 on past dates94%Held-outHeld-out, draw 1: 47.5% of K = 97.29 on past dates47%Held-out, draw 2: 21.0% of K = 93.62 on past dates21%at a tie: no reference answerRegulatory near-tieat a tie: no reference answerat a tie: no reference answerat a tie: no reference answermedian 61%0%50%100%
Capital on past observation dates, per main book. On the 14 smooth main books, 21–99% of capital (median 61%) sits on books held on earlier dates, so no sensitivity to today’s positions can account for it. Books at a tie have no reference share. More results →

Limitations and what didn’t work →

Papers

  • Exact Trade-Level Attribution of FRTB-IMA Capital

    arXiv 2609.32238 · 26 September 2026 [arXiv]

  • Auditable FRTB Capital Explanations for AI Agents

    Yuhe Sui

    Workshop on Responsible AI Operations for Finance (RAIOps4Fin), ICAIF 2026 · accepted [News]

  • Which Question Does an FRTB Capital Attribution Answer?

    Yuhe Sui, Yingzhi Tang

    Workshop on Interpretability and Alignment in Financial Models (IAFM), ICAIF 2026 · accepted [News]

QR code for qrsntu.org/projects/frtbtrace/
Scan to open this page