How Digital Twins Can Transform Commercial Bank Risk Pricing Decisions

Commercial banks face increasing pressure from systemic risks such as global economic cycles, interest‑rate fluctuations, and monetary‑policy changes, making traditional cost‑plus or benchmark‑rate pricing methods inadequate. To address this, the authors construct a digital‑twin environment that repeatedly simulates various pricing strategies under different risk scenarios, allowing decision‑makers to select the most profitable approach.

1. Digital‑Twin Construction for Risk Shocks – Thirteen macro‑economic variables (derived from the Federal Reserve stress‑test model and adjusted for China) are identified (see Table 1). Historical crises (1998 Asian, 2008 Global) are mapped to these variables, and their annualized changes are used as inputs for the simulation.

2. Asset‑Side Digital Twin – Four models are built:

Market supply model: multivariate linear regression on regional loan‑type data, yielding 146 sub‑models; 72 with high R², adjusted R², MAPE, and RMSE are retained.

Loan‑increment model: multivariate linear regression with WOE‑encoded variables; 172 potential factors are screened down to 68 (IV), 31 (correlation), 20 (business relevance), and 18 (multicollinearity).

Credit‑risk model: PD & LGD based on multi‑class logistic regression + transition matrix; formulas for PDᵢ, LGDᵢ, EADᵢ₋₁, and cureᵢ are provided.

Early‑repayment model: COX proportional‑hazards regression with stepwise selection, validated by Wald, Log‑Rank tests and C‑Index.

3. Liability‑Side Digital Twin – Also four models:

Market supply model for deposits (4 categories, 112 sub‑models).

Deposit‑increment model mirroring the loan‑increment approach, with factors such as deposit‑price sensitivity and competitor pricing.

Bond‑issuance model: pricing derived from credit‑rating‑linked risk premiums; assumptions include fixed‑coupon bonds and no embedded options (see Fig 3).

Early‑withdrawal model: incorporates customer‑type stickiness and defines sensitivity thresholds (T) for different deposit tenors (see Table 3).

4. Banking Performance Evaluation Model – The digital‑twin outputs the balance‑sheet evolution under each scenario, from which five core indicators are calculated: capital adequacy ratio, non‑performing loan ratio, net interest margin, liquidity coverage ratio, and return on equity. Formulas for each metric are shown in the accompanying figures.

The integrated framework (Fig 1) enables simulation of bank profitability and risk exposure for any combination of macro‑economic shock and pricing policy, supporting regulators and managers in meeting the requirements of the "Commercial Bank Risk Supervision Core Indicators".

5. Conclusion and Future Work – The constructed digital‑twin system can forecast bank performance under systemic risk and guide optimal pricing decisions. The authors suggest extending the platform with deep reinforcement learning to automatically discover the most effective risk‑pricing strategies.