Paper: arXiv 2308.01486
Authors: Rudy Morel, Stéphane Mallat, Jean-Philippe Bouchaud
Abstract
We introduce a Path Shadowing Monte-Carlo method, which provides prediction of future paths, given any generative model. At any given date, it averages future quantities over generated price paths whose past history matches, or shadows’, the actual (observed) history. We test our approach using paths generated from a maximum entropy model of financial prices, based on a recently proposed multi-scale analogue of the standard skewness and kurtosis called Scattering Spectra’. This model promotes diversity of generated paths while reproducing the main statistical properties of financial prices, including stylized facts on volatility roughness. Our method yields state-of-the-art predictions for future realized volatility and allows one to determine conditional option smiles for the S&P500 that outperform both the current version of the Path-Dependent Volatility model and the option market itself.
Complexity vs Empirical Score
- Math Complexity: 8.5/10
- Empirical Rigor: 7.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Research Flowchart
flowchart TD
A["Research Goal:<br>Predict Future Paths & Prices"] --> B{"Key Methodology"}
B --> C["Path Shadowing Monte-Carlo<br>Generate paths matching observed history"]
B --> D["Maximum Entropy Model<br>Using Scattering Spectra"]
C --> E["Computational Process:<br>Monte Carlo Simulation"]
D --> E
E --> F["Key Findings/Outcomes"]
F --> G["State-of-the-art<br>Volatility Prediction"]
F --> H["Superior Option Smiles<br>vs. PDV model & Market"]
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 introduces the Path Shadowing Monte‑Carlo (PS‑MC) method, a way to obtain conditional predictions p(x|xpast) from any generative price model.
Source sentence
Our main contribution is to introduce a new method, that we call Path Shadowing Monte-Carlo (PS-MC), which can be used within any generative model of p(x) to yield a model of p(x|xpast).
Assumes 5
- The real world can be approximated as stationary with fixed, time‑independent parameters. statistical
Source sentence
We first assume that the real world is at least approximately stationary, in the sense that it can be approximated by a statistical model with fixed, time-independent parameters.
- The underlying price model has a continuous density p(x) that is strictly positive everywhere. model
Source sentence
If q is continuous with E{|q(x)|} < +∞ and the distribution p of x is continuous with p(x) > 0 for all x ∈RN, then ...
- Shadowing paths are weighted with a Gaussian kernel based on the distance in the embedding space. model
Source sentence
We choose Gaussian weights, to wit wk = c exp ( -∥h(xk past) −h(expast)∥2 / 2η2 ) with c such that 1/n
- The Scattering Spectra statistics Φ(x) are low‑variance and can be accurately estimated from a single historical realization. statistical
Source sentence
The SS model imposes Epθ{Φ(x)} = Φ(ex), thus for pθ to be a good approximation of the true distribution p, one needs Φ(ex) to be close to the true Ep{Φ(x)}. This amounts to having low-variance statistics Φ.
Show 1 more
- Generated synthetic paths faithfully capture long‑range dependencies needed for predictive power. model
Source sentence
A good model should generate trajectories that capture the long-range dependencies in order for the shadowing paths to have predictive power on the future of expast.
Data
- S&P 500 log‑price series market-prices public
Source sentence
We compute the Scattering Spectra statistics (9) on observed 3772 S&P log‑prices from January 2000 to December 2014, such that all our predictions are out-of-sample.
- Synthetic trajectories from the Scattering Spectra model synthetic unknown
Source sentence
From such statistics we generate 32 768 trajectories of same size 3772 (see Fig. 8 for examples), that represents n ≈115 million paths x1, . . . , xn of size 126+150 days.
Costs
Not modeled
Source sentence
We neglect bid-ask spread and consider the option price CMKT(t, T, K) to be the quoted mid-price.
Finds 2
- The PS‑MC method achieves state‑of‑the‑art predictions for future realized volatility and produces conditional option smiles for the S&P 500 that outperform the current Path‑Dependent Volatility model and the market implied smiles. predictive-accuracy
Source sentence
Our method yields state-of-the-art predictions for future realized volatility and allows one to determine conditional option smiles for the S&P500 that outperform both the current version of the Path-Dependent Volatility model and the option market itself.
- Optimal embedding and weighting parameters (α=1.15, β=0.9, bη=0.075) were calibrated on the generative model itself, showing little change across different volatility horizons due to scale‑invariance. statistical
Source sentence
This yields the following optimal parameters: α = 1.15, β = 0.9, bη = 0.075 and a path embedding hα,β of dimension 34. Let us note that these optimal values barely change when predicting realized variance at different maturities.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 8 related papers). Coverage is partial, so this is neither replication nor refutation.
11 papers in the archive
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.