Paper: arXiv 2411.05257

Authors: Hardik Routray, Bernhard Hientzsch

Abstract

We propose a simple methodology to approximate functions with given asymptotic behavior by specifically constructed terms and an unconstrained deep neural network (DNN). The methodology we describe extends to various asymptotic behaviors and multiple dimensions and is easy to implement. In this work we demonstrate it for linear asymptotic behavior in one-dimensional examples. We apply it to function approximation and regression problems where we measure approximation of only function values (Vanilla Machine Learning’’-VML) or also approximation of function and derivative values (Differential Machine Learning’’-DML) on several examples. We see that enforcing given asymptotic behavior leads to better approximation and faster convergence.

Complexity vs Empirical Score

  • Math Complexity: 8.0/10
  • Empirical Rigor: 6.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper involves advanced mathematics including deep neural networks, splines, and asymptotic analysis, scoring high on math complexity, while it provides concrete implementations and experimental results on financial functions like Black-Scholes, meeting backtest-ready criteria and thus high empirical rigor.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Enforce Asymptotic Behavior<br>in DNNs for Finance"] --> B["Methodology: Construct<br>Specific Terms + Unconstrained DNN"]
  B --> C["Data: 1D Financial Examples<br>Linear Asymptotes"]
  C --> D{"Computational Process"}
  D --> E["Vanilla ML: Approximate<br>Function Values"]
  D --> F["Differential ML: Approximate<br>Function & Derivative Values"]
  E & F --> G["Key Findings:<br>Better Approximation &<br>Faster Convergence"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a simple methodology that enforces a given asymptotic behavior by adding specially constructed terms to an unconstrained deep neural network.

Source sentence
We propose a simple methodology to approximate functions with given asymptotic behavior by specifically constructed terms and an unconstrained deep neural network (DNN).

Assumes 5

  • Pricing functions in quantitative finance are assumed to have linear asymptotes. model
    Source sentence
    In particular pricing functions in quantitative finance often have or are assumed to have linear asymptotes [HS20, Section 2 in Appendix 4].
  • The target function’s asymptotic form is known (or can be estimated) beforehand. other
    Source sentence
    through observation, limiting behavior, or asymptotic methods, we know the asymptotic behavior or form of f
  • Training data can include exact function values and, optionally, exact derivative values. data
    Source sentence
    We are trying to learn a function f from function values (and possibly additional derivative values)
  • Synthetic data are generated from a Black‑Scholes model with fixed parameters. model
    Source sentence
    intermediate time t = 1 year, maturity T = 2 years, volatility σ = 0.1, short rate r = 0% and strike K = 10.
Show 1 more
  • Input space is one‑dimensional and sampled uniformly on [0,20] for the toy examples. data
    Source sentence
    selected x from a uniform distribution between 0 and 20 along with associated true function and derivatives values y and dy/dx.

Data

  • Uniform synthetic input data synthetic unknown
    Source sentence
    selected x from a uniform distribution between 0 and 20 along with associated true function and derivatives values y and dy/dx.
  • Black‑Scholes simulated payoff data synthetic unknown
    Source sentence
    We use X = St and Y = ˜CT . In this test, we use the same setting as in the last section: intermediate time t = 1 year, maturity T = 2 years, volatility σ = 0.1, short rate r = 0% and strike K = 10.

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • Adding asymptotic treatment to VML yields faster and more accurate convergence than VML without it. performance
    Source sentence
    When looking at the difference to the true values, it is evident that VML in conjunction with asymptotic treatment leads to faster and accurate convergence, compared to VML
  • Larger sample sizes improve approximation most noticeably when asymptotic treatment is used for both VML and DML. performance
    Source sentence
    the impact of sample size is most prominent when asymptotic treatment is used in both VML and DML with higher sample size resulting in improved approximation.
  • Introducing a non‑zero interest rate changes coefficients slightly but does not materially affect the results. other
    Source sentence
    Adding r as a drift and in the discounting will change the shape and the coefficients slightly but otherwise does not seem to impact the results.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 36 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.