DIFT.
DIGITAL INNOVATION &
FINANCIAL TECHNOLOGY CONFERENCE
2026–2
CHIANG MAI, THAILAND · 10 OCT 2026
KEYNOTE · PART I / FOUNDATIONS

Quantitative Finance
in the Age of
Digital Innovation

Finance is becoming a quantitative science.

Dr. Pasin MarupanthornVice President, Thai Association of Quantitative Analyst and Financial Engineer (TQFE)
MARKETS
BANKING
RISK
TRADING
DIFT2026–2
ABOUT THE SPEAKER / KEYNOTE
INTRODUCTION · THE SPEAKER

From mathematics to financial markets

Photograph of the speaker supplied in his résumé
YOUR SPEAKERResearch · Teaching
Quantitative Finance

Dr. Pasin Marupanthorn

Lecturer · Quant Researcher · Quant Trader

VICE PRESIDENTThai Association of Quantitative Analyst and Financial Engineer (TQFE)
ACADEMIC & PROFESSIONAL QUALIFICATIONS UK · US · THAILAND
DOCTORATE
Ph.D. Actuarial Mathematics
Heriot-Watt University · UK
MASTER'S DEGREE
M.Sc. Financial Engineering
WorldQuant University · US
MASTER'S DEGREE
M.Sc. Mathematical Modelling
University of Birmingham · UK
MASTER'S DEGREE
M.Sc. Mathematics
Thammasat University · Thailand
BACHELOR'S DEGREE
B.Sc. Applied Mathematics
Thammasat University · Thailand
PROFESSIONAL QUALIFICATION
CQF
Certificate in Quantitative Finance · CQF Institute
DIFT2026–2
FROM THEORY TO PRACTICE / KEYNOTE
INTRODUCTION · BACKGROUND

Bridging research, markets, and practice

A snapshot of quantitative finance topics from selected seminars and professional forums.

MATHEMATICAL THEORY→QUANTITATIVE MODELS→FINANCIAL APPLICATIONS
DIFT2026–2
QUANTITATIVE FINANCE / PART I
01 · FOUNDATION

What is quantitative finance?

“

The application of mathematics, probability, statistics, and computation to value assets, understand markets, manage risk, and make financial decisions under uncertainty.

∑

Mathematics

Models & optimization

σ

Statistics

Inference & uncertainty

{ }

Computer science

Algorithms & simulation

↗

Economics & finance

Markets & incentives

MODELS→MEASUREMENTS→FINANCIAL DECISIONS
DIFT2026–2
QUANTITATIVE FINANCE / PART I
01 · FOUNDATION

Five questions that organize the field

01

WHAT IS IT WORTH?

Asset pricing & valuation

↗
02

WHAT COULD GO WRONG?

Financial risk management

↗
03

WHERE SHOULD CAPITAL GO?

Portfolio optimization

↗
04

WHEN AND HOW TO TRADE?

Quantitative trading

↗
05

WHOM SHOULD WE LEND TO?

Credit & banking analytics

↗
BETTER
FINANCIAL
DECISIONS

Different problems.
One quantitative language.

DIFT2026–2
QUANTITATIVE FINANCE / PART I
01 · FOUNDATION

Finance means deciding under uncertainty

ONE PRESENT, MANY POSSIBLE FUTURESSCHEMATIC
TODAYFUTURE ASSET VALUE Illustrative central path Possible outcomes widen over time
THE CENTRAL IDEA

A forecast is
not a guarantee.

Good decisions require more than a single prediction.

01Expected outcomes
02Ranges of uncertainty
03Adverse scenarios
04Decision constraints
DIFT2026–2
QUANTITATIVE FINANCE / PART I
01 · FOUNDATION

The mathematical toolkit

P

Probability & stochastic processes

Random asset prices and rates

σ

Statistics & econometrics

Estimate and test models

∫

Calculus & differential equations

Dynamic financial systems

min

Optimization & operations research

Allocate limited capital

MC

Numerical methods & Monte Carlo

Solve difficult models

AI

Machine learning & computation

Patterns and prediction

ONE FINANCIAL PROBLEM→MULTIPLE QUANTITATIVE METHODS
DIFT2026–2
QUANTITATIVE FINANCE / PART I
01 · SUBJECTS & SKILLS

What subjects does a quant study?

The subjects form a learning ladder: foundations → modeling → real-world decisions.
01FOUNDATION

Mathematics & probability

CalculusSensitivities and changing quantities
Linear algebraFactors, covariance and portfolios
Probability theoryRandom events and conditional risk
Statistical inferenceParameters, estimates and tests
02DYNAMIC MODELS

Stochastic & statistical models

Stochastic processes / calculusUncertain prices, rates and defaults
Differential equationsPricing and financial dynamics
Time series / econometricsPatterns, volatility and dependence
Numerical methodsSolve models without closed forms
03DECISION SCIENCE

Computation & intelligence

OptimizationAllocate capital under constraints
Monte Carlo simulationGenerate distributions and scenarios
Machine learningPatterns, prediction and adaptation
Validation & backtestingTest robustness beyond training data
FINANCE SUBJECTS: Financial markets · Economics · Risk management · Instruments12 QUANT SUBJECTS ↗
DIFT2026–2
QUANTITATIVE FINANCE / PART I
01 · PROGRAMMING & DATA

Programming makes models usable

A quant workflow needs more than one programming language.
01 · QUANT WORKFLOW

ACQUIRE & CLEAN

Reliable, timestamped market and institutional data.

APIsSQLData pipelines
02 · QUANT WORKFLOW

RESEARCH & TEST

Estimate, simulate, optimize and backtest financial models.

PythonRMonte Carlo
03 · QUANT WORKFLOW

DEPLOY & MONITOR

Run controlled systems with tests, versioning and monitoring.

C++ / PythonGitProduction checks
Py

Python

Research, pricing, data science

C++

C++

Speed, simulation, execution

R

R

Statistics, econometrics, insurance

SQL

SQL

Market and client data queries

DIFT2026–2
QUANTITATIVE FINANCE / PART I
02 · MAP OF THE FIELD

The landscape of quantitative finance

QUANTITATIVE FINANCE
01

VALUATION & ENGINEERING

Asset pricing · Derivatives · Fixed income

↗
02

INVESTMENT & TRADING

Portfolio management · Quant trading

◈
03

RISK & INSTITUTIONS

Market risk · Credit risk · Insurance · ALM

◇
04

FINANCIAL MARKETS

Market microstructure · Execution · Liquidity

⇄
05

CROSS-CUTTING QUANTITATIVE METHODS

Financial econometrics · Numerical finance · Optimization · Machine learning · Systemic risk

These are overlapping specializations, not rigid boundaries.
DIFT2026–2
QUANTITATIVE FINANCE / PART I
02 · PROFESSIONAL SPECIALIZATIONS

What kind of quant are you?

One way to classify a quant is by their job: which financial decision do they support?
∂
01 · VALUATION / STRUCTURING

Pricing & structuring

Derivatives pricingRates & FX modelsFinancial engineering
Σ
02 · ASSET MANAGEMENT

Investment & allocation

Portfolio optimizationFactor investingSystematic strategies
σ
03 · INSTITUTIONAL RISK

Risk, credit & capital

Market / credit riskInsurance & ALMStress testing
↔
04 · TRADING INFRASTRUCTURE

Trading & liquidity

Execution algorithmsMarket microstructureMarket making
CROSS-CUTTING SKILLS: Econometrics · Monte Carlo · Optimization · Machine learning · Model validation
DIFT2026–2
QUANTITATIVE FINANCE / PART I
02 · ASSET-CLASS SPECIALIZATIONS

Quant by asset class: eight markets

Another way to classify a quant is by the market, instrument and risk factors they model.
ASSET 01CORE

Equities

Stocks, indices, equity options

Volatility · factors
ASSET 02CORE

Rates & fixed income

Bonds, interest-rate swaps

Yield curves · duration
ASSET 03CORE

Foreign exchange

Spot FX, forwards, FX options

FX hedges · carry
ASSET 04CORE

Commodities

Energy, metals, agriculture

Futures · seasonality
ASSET 05CORE

Credit

Corporate bonds, CDS

Default · credit spreads
ASSET 06EMERGING

Digital assets / DeFi

Tokens, crypto options, AMMs

On-chain liquidity
ASSET 07SPECIALTY

Carbon & environmental

Emission allowances, offsets

Policy · carbon pricing
ASSET 08SPECIALTY

Insurance-linked

Cat bonds, weather contracts

Event & tail risk
↳Asset class ≠ quant role. An equity quant might price options, allocate portfolios, manage risk or execute orders.
DIFT2026–2
QUANTITATIVE FINANCE / PART I
02 · THE TWO-DIMENSIONAL MAP

What do quants do in each asset class?

The asset is one dimension. The quant function is another. Most desks combine both.
ASSET MARKET
PRICING / STRUCTURING
INVESTMENT / ALPHA
RISK MODELING
TRADING / EXECUTION
Equities
Options & smiles
Factors & portfolios
Beta & tail loss
Execution & liquidity
Rates / bonds
Curves & swaps
Carry & duration
Yield shocks / DV01
Bond execution
FX
Forwards & options
Carry & momentum
Currency exposure
Spot / forward spreads
Commodities
Futures & storage
Term-structure signals
Weather & basis
Spread execution
Credit
Bonds & CDS
Credit selection
PD, LGD & spread
Liquidity & hedging
Digital / DeFi
AMM / options
Cross-venue signals
Liquidation & depeg
On-chain / order book
Carbon
Allowance options
Policy / carry
Compliance & transition
Auctions & liquidity
Insurance-linked
Cat-bond contracts
Cat-risk allocation
Extreme-event losses
ILS liquidity
↔Macro investing and microstructure / market making are strategies or trading scales—not additional asset classes.
DIFT2026–2
QUANTITATIVE FINANCE / PART I
02 · QUANT CAREERS

Quantitative finance careers: many paths, one quantitative mindset

Quantitative Finance Careers infographic
Core roles: Quant Risk Manager · Quant Developer · Financial Engineer · Quant Researcher · Quant Trader · Quant Analyst
Why these careers are attractive: strong industry demand, intellectually challenging work, and often highly competitive compensation.
Shared skill base: probability, statistics, stochastic processes, optimization, Python / C++, machine learning, financial markets, and communication.
DIFT2026–2
QUANTITATIVE FINANCE / PART I
02 · INDUSTRY MOMENTUM

Why is quant finance so popular today?

Quantitative vs traditional hedge funds comparison chart
25%

Quant / systematic strategies now represent a large share of hedge fund assets

The chart highlights the expansion of quant hedge fund AUM over time, showing that systematic strategies have become a major part of the industry.

+3.46%

Recent style-level return leadership

In the comparison shown, systematic equity L/S strategies outperformed fundamental peers in September 2026.

Main reasons quant finance is becoming more popular

✓Data-rich markets: more market, alternative, and digital data than ever before.
✓Computing power: cheaper cloud computing, faster optimization, and scalable simulation.
✓Automation: institutions want systematic, repeatable, and monitorable decision processes.
✓Cross-disciplinary talent: finance now attracts mathematicians, programmers, physicists, and AI specialists.
Interpret carefully: this slide illustrates industry momentum, not a universal claim that quant always outperforms. Performance varies by market regime, costs, model quality, and risk management.
DIFT2026–2
QUANTITATIVE FINANCE / PART I
02 · VALUATION

Asset pricing: What is it worth?

PRESENT VALUE OF FUTURE CASH FLOWS
PV = T∑t=1CFt(1 + r)t
Illustrative model with deterministic cash flows and constant r
NOWPV
t = 1CF1
t = 2CF2
t = 3CF3
t = 4CF4
t = 5CF5
← Future payments are discounted back to today
WHERE IT IS USED
01Stocks & businesses
02Bonds & loans
03Investment projects
04Infrastructure

Value is not identical to the most recent market price.

DIFT2026–2
QUANTITATIVE FINANCE / PART I
02 · FINANCIAL ENGINEERING

Derivatives: Designing financial contracts

EUROPEAN CALL OPTION PAYOFFAT MATURITY
K (strike)UNDERLYING PRICE PAYOFF AT EXPIRY Payoff = Sₜ − K Payoff = 0
CT = max(ST − K, 0)
THE PURPOSE

Transfer, hedge,
or reshape risk.

Underlying assetThe price or event the contract depends on
Contract termsStrike, maturity, payoff, settlement
Typical productsOptions · Futures · Swaps · Structured notes
DIFT2026–2
QUANTITATIVE FINANCE / PART I
02 · INVESTMENT

Portfolio optimization: Where should capital go?

THE EFFICIENT FRONTIERILLUSTRATIVE
Efficient portfolio EFFICIENT FRONTIER EXPECTED RETURN ↑ RISK (VOLATILITY) → Feasible portfolios
THE OPTIMIZATION PROBLEM

Balance return,
risk, and constraints.

minw   wᵀΣw

subject to return and portfolio constraints

Expected returnCovarianceLiquidityLimits
Diversification depends on co-movement, not simply the number of assets.
DIFT2026–2
QUANTITATIVE FINANCE / PART I
02 · RISK

Financial risk: What could go wrong?

LOSS DISTRIBUTION AND TAIL EXPOSURESCHEMATIC
VaR cut-off Worst-loss tail RELATIVE LIKELIHOODLARGER LOSS → Schematic distribution — not empirical data
VaR

Value at Risk

A loss threshold for a chosen confidence level.

ES

Expected Shortfall

Average loss beyond that tail threshold.

ST

Stress testing

Losses under severe but plausible scenarios.

Risk models must remain useful when markets behave abnormally.
DIFT2026–2
QUANTITATIVE FINANCE / PART I
02 · SYSTEMATIC TRADING

Quantitative trading: From signal to execution

01DATA
→
02SIGNAL
→
03RISK
→
04EXECUTE
→
05EVALUATE
TWO ILLUSTRATIVE PRICE PATHSNOT A BACKTEST
HYPOTHETICAL PRICES
WHAT MAKES IT QUANTITATIVE?
01
Defined trading rules
02
Statistical validation
03
Position and risk sizing
04
Execution-aware evaluation
A signal is not a strategy until costs, risks, and implementation are addressed.
DIFT2026–2
QUANTITATIVE FINANCE / PART I
03 · MARKET MECHANISMS

Market microstructure: How trades happen

SIMPLIFIED LIMIT ORDER BOOK PRICES IN THB
BEST BID / BUYERSBEST ASK / SELLERS
PRICEQUANTITY
99.982,400
99.973,150
99.961,800
PRICEQUANTITY
100.022,150
100.033,350
100.041,650
BEST BID99.98
BID–ASK SPREAD0.04
BEST ASK100.02
WHY THIS MATTERS

Execution is part
of the economics.

Liquidity
Transaction costs
Market impact
Price discovery

A correct forecast can still lead to a poor trade at a bad execution price.

DIFT2026–2
QUANTITATIVE FINANCE / PART I
03 · BANKING

Credit risk: Should a bank approve this loan?

SIMPLIFIED EXPECTED CREDIT LOSS
EL = PD × LGD × EAD
PDProbability
of default
LGDLoss given
default
EADExposure
at default

An approximate product formulation under appropriate modeling assumptions.

A LENDING DECISION INVOLVES
01Borrower risk
02Loan pricing
03Capital requirements
04Portfolio exposure
05Ongoing monitoring
APPLIES TO RETAIL · SME · AGRICULTURE
DIFT2026–2
QUANTITATIVE FINANCE / PART I
03 · FIXED INCOME

Interest-rate risk: Managing assets and liabilities

INTEREST-RATE TERM STRUCTURESILLUSTRATIVE
Upward curveInverted curve YIELDMATURITY →

BANK BALANCE SHEET

ASSETSLong-term loansBonds
LIABILITIESDepositsBorrowings

Differences in repricing dates and maturities create interest-rate exposure.

ΔP / P ≈ −Dmod × Δy
First-order bond price sensitivity to a small yield change.
DIFT2026–2
QUANTITATIVE FINANCE / PART I
03 · INSURANCE

Actuarial finance: Pricing uncertain losses

☂

LIFE & HEALTH

Mortality and morbidity risks

≋

AGRICULTURE

Weather, yield and production losses

↯

CATASTROPHE

Floods, storms and extreme events

UNCERTAIN EVENT
→
LOSS MODEL & CONTRACT
→
PREMIUM & CAPITAL
→
FINANCIAL PROTECTION
Risk pooling combines many uncertain exposures.
Parametric insurance links payment to an agreed observable trigger.
DIFT2026–2
QUANTITATIVE FINANCE / PART I
03 · CROSS-CUTTING METHODS

From data to defensible financial decisions

FINANCIAL ECONOMETRICS + COMPUTATIONAL FINANCE
01FINANCIAL DATAPrices · transactions · accounts
→
02ESTIMATEStatistics · econometrics
→
03SIMULATEModels · Monte Carlo
→
04DECIDEValue · allocate · hedge
→
05VALIDATEBacktest · monitor · revise

Financial econometrics

Inference, relationships & uncertainty
Time seriesVolatility modelsBayesian inference

Computational finance

Numerical solutions & implementation
Monte CarloNumerical optimizationMachine learning
✦ A historical fit is not proof of future performance. Out-of-sample validation matters.
DIFT2026–2
QUANTITATIVE FINANCE / PART I
04 · REAL-WORLD INTEGRATION

One financial product. Many quantitative disciplines.

ILLUSTRATIVE CASE STUDY

Climate-sensitive
digital lending

Designing an agricultural loan product for customers exposed to weather and income uncertainty.

Which models are needed to make this product sustainable?
01

CREDIT RISK

Estimate borrower default risk

02

PRODUCT PRICING

Determine sustainable loan terms

03

CAPITAL ALLOCATION

Control the lending portfolio

04

INSURANCE

Transfer selected climate losses

05

ALM & RISK

Stress-test funding and losses

06

DATA & MODELS

Estimate, validate and monitor

DIFT2026–2
QUANTITATIVE FINANCE / PART I
04 · THE BIG PICTURE

Models are powerful. Models are not crystal balls.

✓

WHAT MODELS CAN DO

✓Quantify financial risk
✓Compare alternatives
✓Value complex contracts
✓Support consistent decisions
!

WHAT MODELS CANNOT GUARANTEE

!Predict perfectly
!Prevent regime changes
!Eliminate every risk
!Avoid data and model bias
THE NEXT QUESTION

How will AI, new data, and digital platforms transform quantitative finance?

→
DIFT2026–2
QUANTITATIVE FINANCE / PART II
PART II · DIGITAL FINANCE & INSURANCE

Quantitative finance meets the digital financial system

We do not discard asset pricing, portfolio theory, or risk models. We connect them to new data, infrastructure, and decision speeds.
Σ

Quant engine

Models of price, return, uncertainty, and constraints.

+
⌘

Digital layer

AI, APIs, cloud, programmable platforms, and streams.

→
↗

Financial decisions

Faster pricing, allocation, lending, execution, and monitoring.

DIFT2026–2
QUANTITATIVE FINANCE / PART II
THE OPERATING MODEL · DIGITAL FINANCE & INSURANCE

From a model to a live financial decision engine

The continuous decision loop
ObservePrices, flows,
client behavior
→
EstimateRisk, alpha,
default probability
→
DecideOptimization,
policy rules
→
ExecuteOrders, loans,
payments
→
MonitorOutcomes, drift,
retraining
↶ Feedback and governance at every stage
Traditional workflow

Periodic data → analysis → committee decision → reporting

Digital workflow

Streaming data → model → controlled action → continuous feedback

Data quality & permission
Out-of-sample model validity
Execution & conduct controls
DIFT2026–2
QUANTITATIVE FINANCE / PART II
AI & INVESTMENT · DIGITAL FINANCE & INSURANCE

AI predicts patterns. Quants make portfolio decisions.

AI can forecast a signal

PREDICTION

Predict a return distribution, volatility, regime, or factor exposure. A higher forecast score is not the same as an investable strategy.

Asset A
Asset B
Asset C
Asset D
Illustrative predictive scores, not measured returns.

Portfolio construction makes a decision

OPTIMIZATION
maxw   wᵀμ̂ − λwᵀΣw − TC(w)
Forecast return − risk penalty − transaction costs; subject to risk and position constraints.

Three tests before deployment

›Data leakage: was the future visible in training?
›Economic value: does alpha survive turnover and fees?
›Robustness: does it survive regime changes?
DIFT2026–2
QUANTITATIVE FINANCE / PART II
DIGITAL MARKETS · DIGITAL FINANCE & INSURANCE

Quantitative finance inside electronic markets

Electronic limit order book

ILLUSTRATIVE
BIDS · buyers
100.00 1,200 shares
99.99 2,800 shares
99.98 4,100 shares
ASKS · sellers
100.02 900 shares
100.03 2,200 shares
100.04 3,900 shares
Bid–ask spread = 100.02 − 100.00 = 0.02
Every trade faces liquidity and market-impact constraints.

Why market microstructure matters

›Price discovery: orders continuously update quoted prices.
›Execution quality: signals must survive spreads and fees.
›Latency and depth: speed is useful only with sound controls.
In digital markets, a forecast becomes profit only after a trade can be executed.
DIFT2026–2
QUANTITATIVE FINANCE / PART II
TRADING & EXECUTION · DIGITAL FINANCE & INSURANCE

Execution is an optimization problem, not just a button

The execution trade-off

SCHEMATIC
Cost / riskLonger execution Market impact Price exposure Balance
min   E[execution cost] + λ · execution risk

Quantitative execution strategies

TWAP

Split an order across time.

VWAP

Align participation with market volume.

Adaptive routing

Respond to liquidity across venues.

Even a strong strategy can lose money when implementation costs exceed its predictive edge.
DIFT2026–2
QUANTITATIVE FINANCE / PART II
DIGITAL INVESTMENT · DIGITAL FINANCE & INSURANCE

Robo-advisers: portfolio theory delivered at scale

Digital wealth management journey
Client goals

Horizon, income, liquidity

→
Risk budget

Capacity & preferences

→
Portfolio

Constrained allocation

→
Implement

Rebalance, taxes, costs

→
Review

Drift and suitability

Stocks · 42%Bonds · 18%Cash · 26%Alternatives · 14%
Illustrative allocation only—not a recommendation.

Quantitative finance behind the app

›Goal-aware portfolio construction
›Tax-aware / cost-aware rebalancing
›Suitability, explainability and investor protections
DIFT2026–2
QUANTITATIVE FINANCE / PART II
FINANCIAL ENGINEERING · DIGITAL FINANCE & INSURANCE

Digital derivatives: live pricing, Greeks and hedging

Live derivatives risk dashboard

ILLUSTRATIVE
DELTA · price sensitivity+0.62per unit of underlying move
GAMMA · delta curvature+0.034units per price point
VEGA · volatility sensitivity+12.4value units per volatility point, convention-dependent
ΔV ≈ Δ · ΔS + ½ Γ(ΔS)² + ν · Δσ
①

Price

Calibrate market and model inputs.

②

Hedge

Translate Greeks into risk-adjusting trades.

③

Monitor

Update limits, sensitivities, liquidity, and model drift.

DIFT2026–2
QUANTITATIVE FINANCE / PART II
DIGITAL BANKING · DIGITAL FINANCE & INSURANCE

Digital credit: from probability of default to loan decisions

A responsible lending pipeline
Consented data

Transactions, income, bureau

→
Credit model

PD, uncertainty, drift

→
Decision

Limits, affordability, price

→
Monitoring

Repayment and resilience

Financial objective

QUANT CORE
Expected loss = PD × LGD × EAD
Use credit risk estimates alongside funding costs, affordability, capital and lending-policy constraints.

What responsible automation adds

GOVERNANCE
›Consent, privacy, and data minimization
›Fairness checks and reason codes
›Human review and model-drift triggers
DIFT2026–2
QUANTITATIVE FINANCE / PART II
TOKENIZED MARKETS · DIGITAL FINANCE & INSURANCE

Tokenisation: new financial plumbing, familiar economics

Traditional market plumbing

SEQUENTIAL
Trade

Match instructions

→
Reconcile

Compare books

→
Settle

Transfer assets

Multiple ledgers and intermediaries may require coordination.

Tokenised infrastructure

PROGRAMMABLE
Delivery versus payment
Asset transfer ⇄ payment settlement

Linked execution can reduce reconciliation and selected counterparty or settlement frictions.

Quant still prices

Claims, discount rates, options and default risks do not disappear.

Technology changes the workflow

Ownership records and contingent settlement actions become programmable.

New operating risks remain

Custody, security, interoperability, and legal finality matter.

Reference: BIS Annual Economic Report 2025, Chapter III; BIS Annual Economic Report 2026.
DIFT2026–2
QUANTITATIVE FINANCE / PART II
DECENTRALIZED FINANCE · DIGITAL FINANCE & INSURANCE

DeFi: mathematical market making with AMMs

A constant-product automated market maker

SIMPLIFIED
x · y = k
Reserve yReserve x Trade changes reserves
The curve illustrates a classical two-asset AMM; fees and execution mechanics are omitted.

Financial questions for DeFi quants

›Pricing: how does the pool set marginal exchange rates?
›Liquidity: what happens when a trade is large relative to reserves?
›Risk: arbitrage, impermanent loss, oracle and smart-contract risk?
DeFi replaces some intermediaries with algorithms. It does not remove financial risk.
References: Uniswap v2 documentation; BIS DeFi risks.
DIFT2026–2
QUANTITATIVE FINANCE / PART II
DIGITAL PAYMENTS · DIGITAL FINANCE & INSURANCE

Payments: fraud analytics, liquidity and settlement

Quantitative risk management inside payments
Initiate

Customer & merchant

→
Risk score

Fraud / AML models

→
Authorize

Controls & friction

→
Settle

Finality & liquidity

A decision threshold balances errors

min   fraud loss + false-decline cost + operating friction

The lowest fraud rate alone is not the whole objective.

Quant risk questions

›How much fraud can be prevented without rejecting legitimate users?
›How much settlement liquidity is required?
›How does real-time operation affect resilience?
DIFT2026–2
QUANTITATIVE FINANCE / PART II
DIGITAL RISK · DIGITAL FINANCE & INSURANCE

Digital finance needs stronger quantitative model governance

△
Model risk and drift
Backtests fail when relationships or data distributions change.
⊛
Shared-model concentration
Similar AI systems may amplify correlated decisions.
⌘
Cyber and third-party risk
Dependencies create new points of failure.
≡
Fairness and explainability
Opaque decisions need transparent controls and recourse.

Four governance gates

EVERY MODEL

Validate

Independent testing

Constrain

Limits & overrides

Monitor

Performance & drift

Govern

Accountability & audit

Reference: Financial Stability Board, The Financial Stability Implications of Artificial Intelligence (2024).
DIFT2026–2
QUANTITATIVE FINANCE / PART II
DIGITAL INSURANCE · DIGITAL FINANCE & INSURANCE

Insurance: AI underwriting and parametric protection

Parametric insurance: model the trigger

ILLUSTRATIVE
Payout Trigger KObserved weather index Payout LNo payout
Payout = L · 1{Index ≥ K}

Digital insurance needs three disciplines

›Actuarial model: price the trigger probability and loss distribution.
›Reliable data: weather stations, satellite or IoT observations.
›Smart delivery: automate verified claims and controlled payouts.
Basis risk: an insured customer may experience a real loss without triggering a payout—or vice versa.
Reference: IAIS AI supervision application paper (2025).
DIFT2026–2
QUANTITATIVE FINANCE / PART II
INTEGRATED APPLICATION · DIGITAL FINANCE & INSURANCE

Digital credit meets climate protection

Illustrative Thai agricultural finance service
Digital data

Consented cashflow + weather

→
Credit risk

PD / LGD and uncertainty

→
Loan terms

Pricing & capital limits

→
Protection

Optional parametric cover

→
Monitoring

Repayment, triggers, fairness

Σ

Quantitative finance

Credit, pricing, portfolio limits and capital models.

⌘

Digital technology

Connected services, data pipelines and automated monitoring.

✓

Real-world evaluation

Delinquency, fairness, resilience and protection value.

The purpose is not simply a smarter application. It is better financial outcomes under explicit risk constraints.
DIFT2026–2
QUANTITATIVE FINANCE / PART II
CONCLUSION · DIGITAL FINANCE & INSURANCE

The future quant is digital—and accountable

PART II TAKEAWAY
Better mathematics.
Better data and infrastructure.
Better financial decisions.

Tomorrow’s quant is not only a model builder—but a builder of trusted financial decision systems.

Scientific rigor

Sound assumptions, validation, and uncertainty.

Engineering capability

Secure software, scalable data, reliable execution.

Responsibility

Fairness, transparency, model governance and trust.

DIFT2026–2
QUANTITATIVE FINANCE / PART III
PART III · AI IN QUANTITATIVE FINANCE

AI in Quantitative Finance: from prediction to decision

THE THIRD ACT · 20 MINUTES
AI is not a substitute for financial mathematics.
It is a new way to learn, decide, and act under uncertainty.
01Predict

Returns · volatility · credit risk

02Decide

Portfolio weights · trades · hedges

03Validate

Out-of-sample value · costs · governance

DIFT2026–2
QUANTITATIVE FINANCE / PART III
METHODS · FORECASTS / ACTIONS / LANGUAGE

The AI toolkit: three families, three jobs

SUPERVISED LEARNING

Learn a conditional forecast

Regression · classification · boosted trees · neural networks

xt → fθ(xt) → μ̂t+1

Typical output: expected return, volatility, default probability.

REINFORCEMENT LEARNING

Learn a decision policy

Actions · feedback · risk-sensitive rewards · constraints

st → πθ(st) → at

Typical output: trade, hedge or execution decision.

GENERATIVE / LANGUAGE AI

Process and generate information

Documents · retrieval · embeddings · scenario generation

text / data → representation

Typical output: structured events, research support, simulations.

Critical distinction: accurate statistical predictions are only inputs. The financial objective still includes risk, trading costs, liquidity and constraints.
DIFT2026–2
QUANTITATIVE FINANCE / PART III
ASSET PRICING · FINANCIAL ECONOMETRICS

Machine learning for expected returns and alpha

A cross-sectional return forecasting engine

Inputs at tMomentum · value · liquidity · macro
→
ML modelNonlinear interactions and regularization
→
Predicted alphaRank securities; estimate uncertainty
μ̂i,t+1 = fθ(Xi,t, Mt)
Finance test: Is the signal stable across periods, markets and transaction-cost assumptions?

Examples of predictor interactions

Model sensitivity (illustrative)Market characteristicsNonlinear feature effects

Classical linear benchmark vs. flexible ML pattern. Diagram is conceptual, not empirical performance.

Gu, Kelly & Xiu (2020), Review of Financial Studies.

DIFT2026–2
QUANTITATIVE FINANCE / PART III
TIME SERIES · DYNAMIC RISK STATES

AI can learn changing volatility and market regimes

When the data-generating process changes

Normal conditionsStress regimeTime →Illustrative realized volatility

Learning nonlinear time dependence is useful only if regime changes can be detected and managed out of sample.

Typical quantitative problems

◆
Forecast volatility: GARCH, ML, recurrent / attention models.
◆
Detect regimes: hidden-state and probabilistic classifiers.
◆
Model dependence: joint tails, co-movement, stress transmission.
P(St+1 = j | 𝓕t)

Interpret probabilities as model outputs, not knowledge of future regimes.

DIFT2026–2
QUANTITATIVE FINANCE / PART III
LLMs · FINANCIAL DOCUMENTS / EVENTS

Financial language models: from text to signals

Extract structured information, then test its financial relevance

Public documentsEarnings reports · filings · central-bank releases
→
Financial NLPEvent extraction · sentiment · embeddings
→
Time-stamped signalSurprises · risk flags · portfolio features
→
Out-of-sample testEconomic effects net of realistic frictions
AS-OF DISCIPLINE
If news is released at 08:30, the model may use it after 08:30—not at yesterday’s close.
FAILURE MODES
Hallucinated figures, wrong entity mapping, stale revisions and prompt injection.
Rule: the language model is an information-processing component; market data provenance and event time determine whether its signal is usable.
DIFT2026–2
QUANTITATIVE FINANCE / PART III
ASSET MANAGEMENT · CAPITAL ALLOCATION

AI predictions meet portfolio optimization

Forecasts are uncertain inputs, not portfolio weights

AI forecastsExpected return μ̂
+
Risk estimateCovariance Σ̂
→
OptimizerWeights, trades and limits
maxw { μ̂ᵀw − λwᵀΣ̂w − κ∥w − wold∥1 }

Risk and turnover

Penalize volatility and costly rebalancing.

Real constraints

Budget · exposure · liquidity · leverage.

Practical system architecture

① Predictive modelExpected return and confidence intervals
② Robust portfolio engineConcentration limits + turnover controls
③ Execution and monitoringNet P&L, risk attribution, drift detection
Objective: financially useful decisions, not perfect point forecasts.
DIFT2026–2
QUANTITATIVE FINANCE / PART III
ALGORITHMIC TRADING · OPTIMAL EXECUTION

Reinforcement learning: trading as sequential control

What RL learns: an action policy

Market stateSpread · depth · inventory
→
Policy πθChoose an action
→
ExecutionPost · cross · wait
Rewardt = ΔVt − Costt − λ Riskt
Important: never let a trading agent explore unrestricted in live markets.

Action space for a buy order

Limit order

Lower fees / adverse selection risk.

Market order

Faster fill / potential impact.

Wait

Avoid immediate cost / timing risk.

Reduce size

Lower inventory and footprint.

Test in simulators and historical replay. Compare with TWAP, VWAP and implementation-shortfall baselines; simulation gaps matter.

DIFT2026–2
QUANTITATIVE FINANCE / PART III
DERIVATIVES · NONLINEAR HEDGING

Deep hedging: AI learns how to manage option risk

Traditional approach

MODEL-BASED DELTA HEDGE
Δ = ∂V / ∂S
›Estimate Greeks under a pricing model.
›Rebalance as prices and sensitivities change.
›Transaction costs make continual rehedging costly.

Learned hedging policy

DEEP HEDGING / RL
minπ ρ(Lπ) + E[Cπ]
›Inputs: market state and risk positions.
›Actions: hedges that meet constraints.
›Objective: loss risk plus trading frictions.
Industry relevance: transaction costs, liquidity constraints and multiple instruments make hedging a realistic learning problem. See Buehler et al. (2019).
DIFT2026–2
QUANTITATIVE FINANCE / PART III
GEN AI · MONTE CARLO / SCENARIO DESIGN

Generative models: stress scenarios, not crystal balls

From historical dependence to simulated scenarios

Illustrative scenario cloud Asset 1 return →Asset 2 return →Tail cases

Visual is synthetic and uncalibrated: it does not establish that a model learns extreme losses.

Generative uses in quantitative finance

◆Conditional simulation: draw returns under volatility and macro scenarios.
◆Dependence structure: preserve realistic cross-asset interactions.
◆Stress testing: inject rare shocks that data may underrepresent.
Validation target: tails, autocorrelation, liquidity responses and economic plausibility—not visual realism alone.
DIFT2026–2
QUANTITATIVE FINANCE / PART III
FINANCIAL RISK · QUANTILES / TAILS

AI risk models must forecast the loss distribution

Tail risk: what happens beyond the threshold?

VaRTailSmaller lossLarger loss →Illustrative portfolio-loss density
VaRα = qα(L)   |    ESα = E[L | L ≥ qα(L)]

ES identity shown for continuous loss distributions.

What AI contributes

◆Nonlinear quantile and expected-shortfall forecasts.
◆High-dimensional dependencies and early-warning signals.
◆Scenario-conditioned losses and portfolio stress testing.

What must still be evaluated?

Coverage calibration, exceedance clustering, stress resilience and decision usefulness.

DIFT2026–2
QUANTITATIVE FINANCE / PART III
CROSS-ASSET APPLICATIONS

One AI toolkit—different financial asset classes

THE ECONOMICS AND DATA CHANGE BY MARKET
Asset classTypical inputAI outputQuant decision
EquitiesFactors, news, order bookExpected returns / liquidityStock allocation and execution
Fixed incomeYield curves, macro, flowsCurve and spread dynamicsDuration and hedging
Foreign exchangeRates, order flow, macroConditional FX riskHedge ratio and trade timing
CreditFirm fundamentals, payment dataPD and loss riskLending limits and pricing
CommoditiesWeather, inventories, futuresBasis / term-structure regimesPhysical and financial hedges
Digital assetsTrades, funding, on-chain eventsLiquidity and liquidation riskExposure and execution limits
Not one universal algorithm: the same model family can support very different markets, risks and operating constraints.
DIFT2026–2
QUANTITATIVE FINANCE / PART III
AGENTIC AI · RESEARCH WORKFLOW

Where language-model agents can help a quant team

An assistant may coordinate tools; authority stays with people and controls

Research question“What changed in these issuer filings?”
→
Retrieve evidencePermitted sources and documents
→
Analyze + verifyCitations · code checks · numbers
→
Human approvalDecide whether to act

Useful assistance

Search literature, annotate filings, draft research code.

Needed controls

Source validation, sandboxed tools, permission limits.

Not autonomous by default

No unattended orders, confidential data leaks or money movement.

Distinction: fluency in language is not the same as verified economic reasoning or trading authorization.
DIFT2026–2
QUANTITATIVE FINANCE / PART III
SCIENTIFIC VALIDATION · NO LOOK-AHEAD

Backtesting: the real exam for AI in finance

Walk forward in time—never train on tomorrow

TRAIN

Fit parameters on past dates

VALIDATE

Tune only on prior dates

TEST

Untouched forward period

ROLL

Refit on newly available history

Past Future →

Information timing

Point-in-time data and actual publication delays.

Label overlap

Purge / embargo where overlapping windows contaminate splits.

Market frictions

Fees, spread, slippage, impact and financing.

Research selection

Account for many trials, false discovery and drift.

Net strategy P&L = Gross P&L − Fees − Spread − Impact − Financing
Central rule: A backtest should reproduce the information, constraints and costs a real investment decision would have faced.
DIFT2026–2
QUANTITATIVE FINANCE / PART III
GOVERNANCE · FINANCIAL STABILITY

AI model risk: what could break at scale?

Four risks to manage

Model / data drift

Markets change; calibration and signal strength decay.

Correlated AI behavior

Shared models or signals may intensify crowding.

Provider concentration

Dependence on the same cloud and model infrastructure.

Cyber / manipulation

Data poisoning, prompt injection, fraud and outages.

FSB (2024, 2025); IOSCO (2025). Supervisory approaches are jurisdiction-specific.

A controlled deployment process

01Document data lineage, intended use and model ownership.
02Validate independently against benchmarks and stress tests.
03Apply live limits, monitoring, overrides and incident response.
04Reassess models when markets, vendors or data change.
Trustworthy AI in finance requires lifecycle governance—not only an accuracy metric.
DIFT2026–2
QUANTITATIVE FINANCE / PART III
CONCLUSION · THE COMPLETE QUANT STACK

The future quant: AI + financial science + accountability

PART III TAKEAWAY
Financial AI succeeds when it connects
reliable evidence to risk-aware decisions.
Economic questionWhat is the real decision and its payoff?
→
AI and mathematicsModels, optimization, uncertainty
→
Real executionCosts, capital, liquidity, constraints
→
AccountabilityValidation, people, governance
For the next generation of quants: learn mathematical finance, statistical ML, software engineering and responsible deployment—not just how to prompt a model.
DIFT2026–2
QUANTITATIVE FINANCE / RESEARCH SPOTLIGHT
PART IV · OUR RESEARCH · 15 MINUTES

Research spotlight: Hybrid stablecoin lending

FROM QUANTITATIVE FINANCE TO DECENTRALIZED FINANCE
Can a lending-rate policy stabilize a stablecoin before liquidity disappears?
Equilibrium, Liquidity Spirals and First-Passage Control in Hybrid Stablecoin Lending
Nassamon Bootwisas · Pasin MarupanthornTheory + synthetic simulationAudited revision · October 2026

01 · System feedback

Peg, lending liquidity and trading interact.

02 · Stability

How strong and how fast should rates respond?

03 · First passage

Can policy reduce the chance of a real depeg event?

DIFT2026–2
QUANTITATIVE FINANCE / RESEARCH SPOTLIGHT
THE FINANCIAL QUESTION · REINFORCING FEEDBACK

Why a small depeg can trigger a liquidity spiral

A NEGATIVE FEEDBACK LOOP CAN TURN POSITIVE
Stablecoin sells offThe peg price falls below its target.
→
Scarcity risesWithdrawals and leverage stress reserves.
→
Borrowing rates reactPremium may relieve or amplify pressure.
→
Trading changesTransactions move the market price.
↻
Reserve liquidity

Cash available inside the lending pool to satisfy withdrawals and new borrowing.

Market liquidity

Trading depth available for buying or selling the stablecoin in external venues.

Research question: Can a rate response simultaneously improve the peg, preserve lending liquidity and avoid destabilizing feedback?
DIFT2026–2
QUANTITATIVE FINANCE / RESEARCH SPOTLIGHT
MECHANISM · ACCOUNTS + AMM + EVENT CLOCKS

A hybrid model links lending accounts to the market price

LENDING RESERVE · ACCOUNTING
Cash, debt and utilization
A = C + B   ·   U = B/(C + B)
CCash C: changes with borrowing, cash repayment, deposits and withdrawals.
BDebt B: changes with borrowing, repayments, liquidations and accrual.
EXTERNAL MARKET · CONSTANT-PRODUCT VENUE
Stablecoin price and trading depth
P = e−x   ·   H V = const.
PPrice P: a larger log-discount x means a weaker peg.
ΔTrades: buying or selling stablecoins generates finite price jumps.
RESERVE EVENT CHANNELS
BorrowCash repaySupplyWithdrawaToken repayLiquidate

Events create discontinuous cash/debt and market changes; oracle observations and rate quotes update on their own schedules.

DIFT2026–2
QUANTITATIVE FINANCE / RESEARCH SPOTLIGHT
CONTROL DESIGN · A PROPOSAL, NOT A DEPLOYED FEATURE

The proposed controller adds a gated peg-sensitive premium

ADDITIONAL BORROWING-RATE PREMIUM
Π(v,S) = G(U) · min{r̄, k(S − s₀)+}
1Utilization gate G(U): turns on as the reserve becomes scarce.
2Peg signal S: filtered observed log-discount, beyond deadband s₀.
3Gain k and cap r̄: sensitivity and maximum extra annual rate.
CONCEPTUAL UTILIZATION GATE
10u₀u₁U
Key distinction: the peg-sensitive premium is proposed in this paper. It is not presented as an existing Aave utilization-rate rule.
DIFT2026–2
QUANTITATIVE FINANCE / RESEARCH SPOTLIGHT
TRACTABLE STRESS CLOSURE · A DYNAMIC SYSTEM

Three risk coordinates reveal the feedback mechanism

SCARCITY
v: reserve pressure
dv = (ℓ − av + bx − ηr)dt + dMᵛ

A peg discount can amplify liquidity stress.

DEPEG
x: price discount
dx = (m + cv − dx − χr)dt + dMˣ

Reserve pressure can generate selling or lost peg support.

INFORMATION
S: observed signal
dS = ω(x+ − S)dt

The filter does not observe or react instantaneously.

v · ScarcityCash and debt stress
⇄
x · Peg discountExternal price pressure
→
S · FilterDelayed information
→
r · PremiumPolicy response
Interpretation: this low-dimensional closure freezes local compensator responses. It is not a general identity for every accounting-level transaction intensity.
DIFT2026–2
QUANTITATIVE FINANCE / RESEARCH SPOTLIGHT
THEOREM 1 · ACTIVE-BRANCH EQUILIBRIUM AND STABILITY

Stronger feedback is not always safer: a finite gain window

Original research figure: largest real eigenvalue versus peg-feedback gain, showing a finite stability window
EXACT SPECTRAL STABILITY · SYNTHETIC EXAMPLE
3/13 < k < 319/49

Approximately 0.231 < k < 6.510.

ADDITIONAL CAP / BRANCH TEST
k > 6/11 ≈ 0.545

Needed here for the active, unsaturated equilibrium with the illustrative 200-bp cap. Gate consistency is also required.

Takeaway: too little feedback fails to stabilize; sufficiently delayed, overly strong feedback can also become unstable.
Source: manuscript Theorem 1 and gain-window figure; the bounds are parameter-specific, not protocol settings.
DIFT2026–2
QUANTITATIVE FINANCE / RESEARCH SPOTLIGHT
THEOREM 1 · A FINITE-CAP ESCAPE REGION

A capped rate cannot rescue every liquidity spiral

A SUPERCRITICAL POSITIVE FEEDBACK LOOP
b c > a d
If the cross-effects between scarcity and peg discount exceed their natural mean reversion, the deterministic closure has an unstable mode.
Scarcity v ↑Less reserve liquidity
⇄
Discount x ↑Greater depeg pressure
CONDITIONAL FINITE-CAP RESULT
wᵀ(v₀,x₀) > r̄·wᵀ(η,χ) / λ₊
Above this initial stress threshold, a premium bounded by the cap cannot prevent divergence in the specified deterministic closure.
Practical interpretation

Some states require liquidity or balance-sheet intervention; merely raising the borrowing rate is insufficient.

Assumes b c > a d, nonnegative loads, χ ≥ 0 and the positive unstable eigenmode of the stated closure.

DIFT2026–2
QUANTITATIVE FINANCE / RESEARCH SPOTLIGHT
THEOREM 2 · SAMPLED STABILITY AND REAL INFORMATION

Oracle delays sharply reduce the available control window

Original research figure comparing continuous, held, and stale observations of the stablecoin price
FIRST COMPUTED LOSS OF SAMPLED STABILITY · k = 3
13.969 h

Continuously observed peg input

6.631 h

Current observation held between updates

3.372 h

One-epoch stale observation

Implementation matters: a stable continuous-time model can become unstable under slower or stale quotes.
These are first crossing roots for the study's synthetic parameters, not universal safe update intervals or real oracle requirements.
DIFT2026–2
QUANTITATIVE FINANCE / RESEARCH SPOTLIGHT
THEOREM 3 · CONTINUOUS-TIME FIRST-PASSAGE RISK

First-passage risk counts every depeg or liquidity hit

DEFINE DISTINCT FAILURE EVENTS
Depeg
FP = {x ≥ −log(1 − δ)}
Liquidity stress
FU = {U ≥ uc}
Joint stress is the intersection

FJ = FP ∩ FU. The union means either event and is a different risk metric.

KILLED-PROCESS VIEW OF DEFAULT / DEPEG Failure barrierJump overshoot = hitTime
q(t,z) = 1 − QT−t1(z)
A first-passage probability counts any crossing before the horizon—including a jump over the boundary or a hit between observation times.
DIFT2026–2
QUANTITATIVE FINANCE / RESEARCH SPOTLIGHT
THEOREM 4 · BELLMAN RECURSION AND BASELINE CERTIFICATE

Risk-aware sampled control has a conditional guarantee

OPTIMIZE: HIT PROBABILITY + NONNEGATIVE PREMIUM BURDEN
Vn(z) = minr ∈ A(z) { Hnr(z) + Cnr(z) + Qnr Vn+1(z) }
H = immediate hit risk

Probability of crossing a failure threshold before the next decision.

C = policy burden

Nonnegative cost of charging an additional premium.

Q = safe continuation

Probability-weighted future risk on surviving paths.

BASELINE IS A FEASIBLE CANDIDATE
Optimal risk + burden ≤ Base failure risk

The Bellman claim is for the specified transition law, candidate actions and decision schedule.

NON-DETERIORATION CERTIFICATE
Dn ≥ 0 on every live state

A nonnegative baseline-risk residual yields a weak guarantee; strict improvement needs positive visited residual.

DIFT2026–2
QUANTITATIVE FINANCE / RESEARCH SPOTLIGHT
SEVEN-DAY SCALAR BENCHMARK · FINITE-STATE MODEL

Synthetic results: less depeg risk, with an economic tradeoff

MODEL: 601 TRANSIENT STATES · 7-DAY FIRST PASSAGE
Probability of a depeg event
Base policy24.59%
Preset gate13.24%
Risk-aware quote13.90%
Optimal does not mean lowest raw risk. The Bellman policy balances failure risk with a penalty for additional borrowing rates.

Risk-aware policy burden: 0.03033; total modeled objective: 0.16932. Burden is not a dollar cost without exposure conversion.

Original research figure: optimal sampled annual premium quotes depend on observed depeg and time remaining
DIFT2026–2
QUANTITATIVE FINANCE / RESEARCH SPOTLIGHT
ACCOUNTING-LEVEL SIMULATION · 4,000 PAIRED PATHS PER SCENARIO

Hybrid simulation: protection depends on the scenario

SEVEN-DAY JOINT STRESS · DEPEG AND HIGH UTILIZATION
Synthetic experiments only
Aligned rate response
Base
22.40%
Gate
21.12%
Δ −1.275 pp
Paired 95% CI: [−1.649, −0.901] pp
No response
Base
25.57%
Gate
25.57%
Δ 0.000 pp
No observed difference in this scenario
Adverse supply response
Base
24.02%
Gate
24.12%
Δ +0.100 pp
Paired 95% CI: [−0.119, 0.319] pp
STANDALONE DEPEG RISK
Approximately 42.2% across base and gate arms in all three scenarios.
INTERPRETATION
A preset premium can help, do nothing or worsen joint stress depending on the modeled response.
Do not overclaim: scenario estimates are not empirical outperformance; the preset gate is not the solved Bellman policy for the accounting-level model.
DIFT2026–2
QUANTITATIVE FINANCE / RESEARCH SPOTLIGHT
PRACTICAL TAKEAWAYS · VERIFICATION BOUNDARIES

From mathematical insight to responsible DeFi risk control

01 · GAIN
Tune the feedback, do not maximize it

Check active stability, the utilization gate and cap consistency separately.

02 · LATENCY
Validate the actual oracle and quote clock

Held and stale information can change the system's stability boundary.

03 · RISK
Control first-passage events

Measure depeg, liquidity and joint-stress risk separately, including jump overshoots.

04 · EVIDENCE
Calibrate before real deployment

Verify market and transaction responses, model law errors and numerical approximation.

Research boundary: the proofs underwent conventional review and the numerical workflow was reproduced, but all experiments are synthetic. Lean proof candidates were not executed; no machine-verification or deployed-protocol guarantee is claimed.
FINAL MESSAGEA useful stablecoin controller must be stable, informed, risk-sensitive and independently validated—not merely aggressive.
DIFT2026–2
QUANTITATIVE FINANCE / DEFI INSURANCE RESEARCH
PART V · DIGITAL INSURANCE · EMPIRICAL STUDY

Research spotlight: Can you insure a DeFi protocol?

A PUBLIC WINDOW INTO DIGITAL UNDERWRITING
Does an insurance listing tell us anything about a DeFi protocol’s insurability?
Digital Insurance and Protocol Insurability in Decentralized Finance: Longitudinal Evidence from Public Cover Listings
EVENTSFirst listings

When a protocol becomes publicly coverable

SCALETVL histories

Lagged economic activity before listing

UNDERWRITINGCover capacity

Available cover capital at snapshot

TERMSAnnual prices

Quoted cover prices, not safety ratings

DIFT2026–2
QUANTITATIVE FINANCE / DEFI INSURANCE RESEARCH
THREE DISTINCT OBSERVABLES · NOT ONE QUALITY SCORE

Digital insurance makes underwriting decisions observable

01
Listing
Can the protocol be named and offered as a public direct-cover product?
MEASURES
Market selection and product availability
02
Capacity
How much potential risk-transfer capital is available for that product?
MEASURES
Current underwriting depth
03
Quoted annual price
At what publicly quoted terms can buyers seek protection?
MEASURES
Pricing, utilization, demand, scarcity
Important distinction: a listed product may have little available capacity; capacity and quoted price do not reveal claims paid or the technical safety of the insured protocol.
DIFT2026–2
QUANTITATIVE FINANCE / DEFI INSURANCE RESEARCH
ECONOMIC LOGIC · COMPETING EXPLANATIONS

Four hypotheses: visibility, maturity, incidents, and capital

H1 · SCALE
Economically material protocols
Higher prior TVL may increase attention, assessability, and cover demand.
Expected positive association
H2 · AGE
New versus established
Long history builds confidence, but newer protocols may attract rapid product expansion.
Sign is ambiguous
H3 · INCIDENTS
Previously hacked
A past exploit can deter underwriting or stimulate demand for protection.
Sign is ambiguous
H4 · CAPITAL
Depth of coverage
Larger protocols may attract more available cover capital among already listed products.
Separate from listing
Identification: lagged variables establish timing, but do not establish why a provider created a listing.
DIFT2026–2
QUANTITATIVE FINANCE / DEFI INSURANCE RESEARCH
REPRODUCIBLE PIPELINE · JULY 2026 DATA CUT

From public events to a longitudinal adoption sample

CURRENT NEXUS PRODUCTS · CONSERVATIVE OBJECT CLASSIFICATION
456products downloaded
→
170public, active, non-private
→
94active direct-protocol products
→
74reviewed linked direct products
31
dated first listings after the observed baseline
25
first-listing events in the primary activity panel
14,284
protocol-months from 854 active protocols
No look-ahead: protocol TVL, age, and public incident history come from the previous month; 45 baseline-covered protocols are left-censored.
Source: manuscript §3 and Table 1. Primary sample requires lagged TVL ≥ US$1 million.
DIFT2026–2
QUANTITATIVE FINANCE / DEFI INSURANCE RESEARCH
DISCRETE-TIME EVENT HISTORY · COMPLEMENTARY LOG-LOG

First listing is modeled as a monthly hazard

DEPENDENT VARIABLE
Yit = 1 in protocol i’s first reviewed direct-cover listing month; 0 in earlier at-risk months.
log[−log(1−hit)] = α + βV log(1+TVLi,t−1)
+ βA log(1+Agei,t−1) + βH Hacki,t−1
+ risk-group + calendar-quarter controls
Model choice: time ordering and protocol-clustered standard errors; logit is a robustness check.
HOW TO READ THE COEFFICIENT
eβ
A hazard ratio compares conditional listing hazards as a predictor changes.
One unit of log TVL
≈ multiplying TVL by 2.72 for large TVL values.
The risk set ends after first listing. This is an association model, not a causal treatment-effect model.
DIFT2026–2
QUANTITATIVE FINANCE / DEFI INSURANCE RESEARCH
MAIN LONGITUDINAL RESULT · ACTIVITY-THRESHOLD TEST

Scale predicts cover listings—but only for active protocols

PAPER FIGURE 2 · HAZARD RATIOS AND ROBUSTNESS
Paper Figure 2: predictors of first public cover listing and TVL threshold sensitivity with 95 percent confidence intervals
PRIMARY TVL EFFECT
1.345×
Monthly listing hazard ratio
95% CI: 1.073–1.687
p = 0.010 · TVL ≥ US$1m
No activity threshold1.106×
p = 0.551 · not statistically reliable
Interpretation: the evidence supports scale-sensitive selection among economically active protocols—not a universal TVL-to-insurability relationship.
DIFT2026–2
QUANTITATIVE FINANCE / DEFI INSURANCE RESEARCH
OTHER LONGITUDINAL ESTIMATES · AVOID SAFETY CLAIMS

Protocol age and prior hacks complicate the story

NEWER PROTOCOLS
0.133
Hazard ratio for log protocol age
p < 0.001. Conditional on scale and calendar quarter, more recently established protocols appear more frequently in new listings.
PRIOR PUBLIC HACK
1.571
Hazard ratio with wide uncertainty
95% CI: 0.172–14.365; p = 0.689. No precise directional evidence.
INTERPRETATION
Selection ≠ Safety
Newer protocols can attract protection demand; observed hacks can either deter capital or increase attention.
Risk group, cohort dynamics, and unobserved decisions still matter.
Do not interpret a lower age hazard ratio as proof that older code is less safe—or the hack estimate as evidence hacks help insurability.
DIFT2026–2
QUANTITATIVE FINANCE / DEFI INSURANCE RESEARCH
EVENT-TIME DESCRIPTION · NOT A TREATMENT EFFECT

Listings often arrive near protocol growth

PAPER FIGURE 1 · LISTING COUNTS AND TVL AROUND FIRST LISTING
Paper Figure 1: monthly reviewed insurance-listing counts and event-time TVL trajectories with interquartile range
MONTH −2
94.7 · TVL index
LISTING MONTH
117.9 · TVL index
MONTH +2
120.3 · TVL index
OCT 2025
7 first listings
DIFT2026–2
QUANTITATIVE FINANCE / DEFI INSURANCE RESEARCH
21 JULY 2026 SNAPSHOT · AVAILABILITY AND MATCHING

Current cover supply: who is actually insurable?

REVIEWED SNAPSHOT
69
Protocols with reviewed active direct cover, compared with 841 supported controls.
61 matched pairs
All post-matching absolute standardized mean differences below 0.10; maximum 0.061.
LOGIT AVERAGE MARGINAL EFFECTS · CURRENT AVAILABILITY
Log TVL
Current economic scale
+0.024
Log chain count
Current breadth across blockchains
+0.037
Reported audit status
Availability of a public audit marker
+0.070
All three associations are statistically significant; this cross-section does not establish that audits or scale cause listing.
Snapshot estimates are distinct from the historical first-listing hazard; contemporaneous attributes should not be backfilled into earlier months.
DIFT2026–2
QUANTITATIVE FINANCE / DEFI INSURANCE RESEARCH
CURRENT PUBLIC COVER · SUPPLY AND PRICE ARE DISTINCT

Bigger protocols receive deeper capacity and lower quotes

PAPER FIGURE 3 · COVER CAPACITY VS. TVL; ADJUSTED AVAILABILITY
Paper Figure 3: protocol total value locked versus public cover capacity, and adjusted logit cover probability with 95 percent confidence band
WITHIN 69 INSURABLE PROTOCOLS
Log capacity regression: log TVL coefficient
+1.080
HC3 robust statistic 3.73
Annual quoted-price regression: log TVL coefficient
−0.539
HC3 robust statistic −4.74
Underwriting depth ≠ technical safety: capacity is available cover supply; quoted price also responds to utilization, demand, product design and capital scarcity.
DIFT2026–2
QUANTITATIVE FINANCE / DEFI INSURANCE RESEARCH
CAUSALITY, MEASUREMENT, AND MARKET BOUNDARIES

What the evidence can—and cannot—establish

STRONGER CLAIMS SUPPORTED
On-chain event timestamps support a documented history of observed first listings.
Lagged TVL is associated with listing among active protocols.
Current capacity, availability, and quoted prices show different patterns.
Reviewed exact product–protocol links reduce ambiguous classification.
CLAIMS NOT SUPPORTED
Insurance listing proves smart-contract safety or will guarantee a claim payment.
Cover causes TVL growth or leads to lower future hack probability.
A quoted annual price is a pure actuarial expected loss estimate.
Single-provider public data capture the complete insurance market.
Key limitations: 25 primary listing events, left-censored history, possible missing failed protocols, public-data revisions, and unobserved purchases, claims and shared-pool constraints.
DIFT2026–2
QUANTITATIVE FINANCE / DEFI INSURANCE RESEARCH
ILLUSTRATIVE DUE-DILIGENCE WORKFLOW · NOT A SAFETY RANKING

How practitioners should use an insurability signal

FOUR QUESTIONS FOR DEFI USERS, TREASURIES, AND RISK TEAMS
01 · IdentifyIs the named insured object the exact protocol, version, or another product?
→
02 · EvaluateWhat are the technical, liquidity, oracle, governance and dependency risks?
→
03 · Read the termsWhat are current available capacity, exclusions, price, and claims conditions?
→
04 · MonitorTrack stress, policy changes, product status and actual protection over time.
USEFUL AS
A screening signal for market relevance, assessability and underwriting attention
NOT VALID AS
A consumer safety score, investment recommendation or guarantee of successful claims
This workflow is a practical interpretation of the study, not an experimentally tested intervention.
DIFT2026–2
QUANTITATIVE FINANCE / DEFI INSURANCE RESEARCH
CONCLUSION · OPEN QUESTIONS · FUTURE RESEARCH

Digital underwriting is more transparent—not risk-free

THE EMPIRICAL MESSAGE
Public cover reveals market selection and capital allocation.
It does not certify protocol safety.
01 · ADOPTIONSelection is conditional

Scale matters primarily among economically active protocols.

02 · CAPITALDepth and quotes differ

Larger covered protocols generally show more available capacity and lower prices.

03 · VALIDITYDo not equate cover with quality

Market attention, demand and risk-based selection can coexist.

Next frontier: multi-provider histories of purchases, capacity, quoted terms, claims, governance decisions and cross-protocol dependencies.