Paper: arXiv 2311.16762
Authors: Matteo Gambara, Giulia Livieri, Andrea Pallavicini
Abstract
Evaluating financial products with early-termination clauses, in particular those with path-dependent structures, is challenging. This paper focuses on Asian options, look-back options, and callable certificates. We will compare regression methods for pricing and computing sensitivities, highlighting modern machine learning techniques against traditional polynomial basis functions. Specifically, we will analyze randomized recurrent and feed-forward neural networks, along with a novel approach using signatures of the underlying price process. For option sensitivities like Delta and Gamma, we will incorporate Chebyshev interpolation. Our findings show that machine learning algorithms often match the accuracy and efficiency of traditional methods for Asian and look-back options, while randomized neural networks are best for callable certificates. Furthermore, we apply Chebyshev interpolation for Delta and Gamma calculations for the first time in Asian options and callable certificates.
Complexity vs Empirical Score
- Math Complexity: 8.5/10
- Empirical Rigor: 8.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced mathematical techniques including randomized recurrent/feed-forward neural networks, signature methods, and Chebyshev interpolation for computing sensitivities, requiring deep theoretical understanding. It demonstrates strong empirical rigor through extensive backtesting against benchmarks (polynomial basis functions, GPR-GHQ algorithm, binomial Markov chain) across Asian, look-back options, and callable certificates, with detailed statistical uncertainty analysis and computational time tables.
Research Flowchart
flowchart TD
A["Research Goal:<br>Compare ML vs. Polynomial Regression<br>for Path-Dependent Options"] --> B["Inputs:<br>Monte Carlo Simulations &<br>Financial Market Data"]
B --> C["Methodology:<br>Randomized Neural Networks<br>& Signature Methods"]
B --> D["Methodology:<br>Chebyshev Interpolation<br>for Greeks Delta & Gamma"]
C --> E["Computational Process:<br>Pricing Asian Options, Look-Back,<br>& Callable Certificates"]
D --> E
E --> F{"Key Findings/Outcomes"}
F --> G1["ML matches accuracy/efficiency<br>for Asian & Look-Back Options"]
F --> G2["Randomized NN best for<br>Callable Certificates"]
F --> G3["First application of Chebyshev<br>Interpolation for Greeks in Asian &<br>Callable Certificates"]
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 compares regression methods for pricing and sensitivities, contrasting modern machine learning techniques with traditional polynomial basis functions and applies Chebyshev interpolation for Delta and Gamma.
Source sentence
We will compare regression methods for pricing and computing sensitivities, highlighting modern machine learning techniques against traditional polynomial basis functions.
Assumes 6
- Asset price dynamics follow the Black‑Scholes model. model
Source sentence
we evaluate the different algorithms by assuming a Black‑Scholes dynamics for underlying assets.
- Only synthetic data are used, not real market data. data
Source sentence
we only consider synthetic data, and not real market data.
- A risk‑neutral pricing measure (Ω,F,Q) with a filtration representing observable market quantities is adopted. market
Source sentence
We choose a risk‑neutral pricing model (Ω, F, Q), with a filtration (Ft)t∈[0,T] representing all the observable market quantities.
- Risk‑free rate, dividend rate and volatility are deterministic constants. model
Source sentence
r is the risk‑free rate, q is the dividend rate, and σ is the volatility, and we assume that all these three quantities are deterministic constants.
Show 2 more
- Options can be exercised on any date in the predefined time grid {TM,…,TN}. execution
Source sentence
We consider that the options can be exercised on any date in the time grid {TM, . . . , TN} previously defined.
- A simple setting for asset price dynamics is preferred over more complex models. other
Source sentence
we think that assuming a simple setting for asset price dynamics is preferable.
Data
- Synthetic Black‑Scholes price paths synthetic unknown
Source sentence
All the experiments are run under a Black‑Scholes dynamics for a single stock price.
Costs
Not applicable
Finds 4
- Machine‑learning algorithms achieve accuracy and computational efficiency comparable to traditional methods for pricing Asian and look‑back options. performance
Source sentence
Our numerical results indicate that machine learning‑based algorithms achieve accuracy and computational efficiency that are comparable to traditional algorithms when pricing Asian and look‑back options.
- Randomized neural networks are the most effective basis functions for pricing callable certificates. performance
Source sentence
we find that randomized neural networks are the most effective choice of basis functions for pricing callable certificates.
- Chebyshev interpolation can be successfully applied for the first time to Delta and Gamma calculations for Asian options and callable certificates. performance
Source sentence
Finally, we discovered that Chebyshev interpolation techniques can be successfully applied, for the first time, to Delta and Gamma calculations for both Asian options and callable certificates.
- Truncated‑signature prices are among the highest for look‑back options and certificates but are matched by R‑FFNNs with much lower computational times. performance
Source sentence
prices obtained from truncated signatures are among the highest for look‑back options and certificates, these are matched by R‑FFNNs which, moreover, have much lower computational times.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 29 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.