Paper: arXiv 2411.12013

Authors: Marco Hening Tallarico, Pablo Olivares

Abstract

This paper studies pricing of weather-derivative (WD) contracts on temperature and precipitation. For temperature-linked strangles in Toronto and Chicago, we benchmark a harmonic-regression/ARMA model against a feed-forward neural network (NN), finding that the NN reduces out-of-sample mean-squared error (MSE) and materially shifts December fair values relative to both the time-series model and the industry-standard Historic Burn Approach (HBA). For precipitation, we employ a compound Poisson–Gamma framework: shape and scale parameters are estimated via maximum likelihood estimation (MLE) and via a convolutional neural network (CNN) trained on 30-day rainfall sequences spanning multiple seasons. The CNN adaptively learns season-specific \((α,β)\) mappings, thereby capturing heterogeneity across regimes that static i.i.d.\ fits miss. At valuation, we assume days are i.i.d.\ \(Γ(\hatα,\hatβ)\) within each regime and apply a mean-count approximation (replacing the Poisson count by its mean (\(n\hatλ\)) to derive closed-form strangle prices. Exploratory analysis of 1981–2023 NASA POWER data confirms pronounced seasonal heterogeneity in \((α,β)\) between summer and winter, demonstrating that static global fits are inadequate. Back-testing on Toronto and Chicago grids shows that our regime-adaptive CNN yields competitive valuations and underscores how model choice can shift strangle prices. Payoffs are evaluated analytically when possible and by simulation elsewhere, enabling a like-for-like comparison of forecasting and valuation methods.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced mathematics, including harmonic regression with ARMA models, compound Poisson–Gamma processes, MLE, and CNN architectures, with significant analytical derivations. It demonstrates strong empirical rigor through extensive back-testing on a 42-year satellite dataset (NASA POWER) for two cities, comparing models against industry benchmarks and reporting specific metrics like MSE and fair value shifts.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Price Weather Derivatives<br>Temperature & Precipitation"] --> B["Data Source:<br>NASA POWER 1981-2023"]
  
  subgraph C ["Temperature Modeling: Toronto & Chicago"]
      B --> C1["Model 1: Time-Series<br>Harmonic Regression + ARMA"]
      B --> C2["Model 2: Neural Network<br>Feed-Forward NN"]
      C1 --> C3["Compare MSE & Fair Values"]
      C2 --> C3
  end

  subgraph D ["Precipitation Modeling"]
      B --> D1["Base Model: Compound Poisson-Gamma<br>MLE Estimation"]
      B --> D2["Adaptive Model: CNN<br>Trained on 30-day Rainfall Sequences"]
      D2 --> D3["Learn Regime-Specific (α, β)<br>Summer vs Winter Heterogeneity"]
  end
  
  C3 --> E["Key Findings<br>NN reduces MSE vs Time-Series<br>Model choice shifts fair values<br>CNN captures precipitation regimes"]
  D3 --> E

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 using satellite-derived NASA POWER data with neural networks for regime‑adaptive pricing of temperature and precipitation weather derivatives.

Source sentence
We are the first to employ the 1981–2023 NASA POWER gridded data set for weather-derivative pricing, providing uniform coverage where gauges are sparse.

Assumes 2

  • Days within each seasonal regime are i.i.d. Gamma(α,β). model
    Source sentence
    At valuation, the approach assumes days are i.i.d. Γ(ˆα, ˆβ) within each seasonal regime and uses a mean-count approximation for analytical tractability
  • Number of wet days follows a Poisson distribution and precipitation on wet days is i.i.d. Gamma. model
    Source sentence
    Let N be the (random) number of wet days in the n-day window and Rk the precipitation on each wet day. We assume N ∼ Pois(λ), Rk | N i.i.d. Γ(α, β)

Data

  • NASA POWER gridded data set · NASA macro public
    1981–2023 · Toronto and Chicago grids
    Source sentence
    We are the first to employ the 1981–2023 NASA POWER gridded data set for weather-derivative pricing, providing uniform coverage where gauges are sparse.

Costs

Not applicable The paper reports no trading or execution result.

Finds 5

  • The feed‑forward neural network reduces out‑of‑sample MSE and materially shifts December strangle fair values versus the harmonic‑regression/ARMA benchmark and the Historic Burn Approach. performance
    Source sentence
    reduces out-of-sample mean-squared error (MSE) and materially shifts December fair values relative to both the time-series model and the industry-standard Historic Burn Approach (HBA).
  • In Toronto for December 2023 the neural network lowers climatological absolute error to 1.76 °C, a 49 % improvement over climatology and 23 percentage‑point gain over the harmonic/ARMA model. predictive-accuracy
    Source sentence
    – Harmonic/ARMA reduces the climatological absolute error from 3.45◦C to 2.54◦C, a 26 % improvement. – The neural network lowers it further to 1.76◦C, yielding a 49 % improvement over climatology and an additional 23 percentage-point gain on the time-series model.
  • In Chicago for December 2023 the neural network achieves 2.97 °C error, a 41 % improvement over climatology and 21 percentage‑point advantage over harmonic/ARMA. predictive-accuracy
    Source sentence
    Chicago (Dec 2023). – Harmonic/ARMA cuts the error from 5.03◦C to 4.01◦C, a 20 % improvement. – The neural network achieves 2.97◦C, corresponding to a 41 % improvement over climatology and a further 21 percentage-point advantage over Harmonic/ARMA.
  • The CNN learns season‑specific (α,β) mappings, capturing heterogeneity across regimes that static i.i.d. fits miss. performance
    Source sentence
    The CNN adaptively learns season-specific $(α,β)$ mappings, thereby capturing heterogeneity across regimes that static i.i.d. fits miss.
  • CNN‑based parameter estimates are less precise than MLE but adapt to seasonal regimes and, combined with the Gamma‑sum formula, yield competitive precipitation contract valuations. pricing-calibration
    Source sentence
    While the CNN’s parameter estimates are less precise, they adapt (α, β) to seasonal regimes—without modeling serial correlation—and (together with the Gamma-sum formula) yield a cl…

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 15 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.

Cited References (100)

  1. Weather Derivatives: Modeling and Pricing Weather-Related Risk (2013)
  2. Deep Learning for Time Series Forecasting Predict the Future with MLPs, CNNs and LSTMs in Python (2018)
  3. An Introduction to Statistical Learning with Applictions in R (2013)
  4. Hands-On Machine Learning with Scikit-Learn, Keras & TensorFlow Concepts, Tools, and Techniques to Build Intelligent Systems (2023)
  5. Stochastic modelling of temperature variations with a view towards weather derivatives (2005)
  6. Option valuation using the fast Fourier transform (1999)
  7. Switching L 'e vy Models in Continuous Time: Finite Distributions and Option Pricing (2005)
  8. A Novel Pricing Method for European Options based on Fourier-Cosine Series Expansions (2008)
and 92 more
  1. Time Series: Theory and Methods, 3rd editon (2016)
  2. Springer Science & Business Media (2008)
  3. Deep Learning for Time Series Forecasting-Predict the Future with MLPs, CNNs and LSTMs in Python (2018)
  4. Pricing Temperature Derivatives under a Time-Changed Levy Model (2020)
  5. A switching mean-reverting Levy model for the temperature dynamic (2022)
  6. Pricing Energy Contracts under Regime Switching Time-Changed models (2020)
  7. Financial Modelling with Jump Processes (2003)
  8. Financial Mathematics: A Comprehensive Treatment (2014)
  9. Knuth: Computers and Typesetting — link
  10. L 'e vy Processes in Finance: Pricing Financial Derivatives (2003)
  11. Continuous-Time Markov Chains — link
  12. Chapter 6: Continuous Time Markov Chains — link
  13. Notes on Memoryless Random Variables — link
  14. A Friendly Introduction to Numerical Analysis (2006)
  15. Stochastic Processes (1996)
  16. Probability Theory and Stochastic Processes with Applications (2009)
  17. Stochastic Processses and their Classification — link
  18. An Introduction to Levy Processes with Applications in Finance (2000)
  19. Subordinators — link
  20. Time-Changed Levy Processes and Option Pricing (2002)
  21. Time Change
  22. Stochastic Differential Equations (1995)
  23. Radon Nokodym Theorems
  24. Computational Methods in Medical Physics (2014)
  25. Inference in Hidden Markov Models (2005)
  26. Mean Likelihood Estimators (2001)
  27. Method of Moments
  28. Gradient Descent - Machine Learning
  29. PDE and Martingale Methods in Option Pricing (2011)
  30. Ito Calculus and Derivative Pricing with Risk-Neutral Measure
  31. A Note on Minimuim Distance Estimates (1979)
  32. Empirical Characteristic Function Estimation and its Applications (2004)
  33. Density and Distribution Estimation
  34. Stochastic Volatility: Modeling and Asymptotic Approaches to Option Pricing and Portfolio Selection (2014)
  35. Martingales and Local Martingales — link
  36. A New Scaling and Squaring Algorithm for the Matrix Exponential (2009)
  37. The Scaling and Squaring Method for the Matrix Exponential Revisited (2005)
  38. Chebyshev and Fourier spectral methods (1989)
  39. Computing a Trust Region Step (1982)
  40. Multinomial Method for Option Pricing under Variance Gamma (2018)
  41. Probability with Martingales (1991)
  42. Modeling electricity prices: Jump diffusion and Regime Switching (2004)
  43. Option pricing in a regime switching stochastic volatility model (2018)
  44. A new approach to the economic analysis of nonstationary time series and the business cycle (1989)
  45. A comparison of regimeswitching temperature modeling approaches for applications in weather derivatives (2014)
  46. Regime switching models on temperature dynamics (2016)
  47. Pricing Rainfall Derivatives at the CME (2013)
  48. Regime-Switching Temperature Dynamics Model for Weather Derivatives (2018)
  49. The Distribution of the Sum of Independent Gamma Random Variables (1985) — link
  50. On the density for sums of independent exponential, Erlang and gamma variates — link
  51. Closed‐form approximated pricing of multivariate derivatives under switching regime models (2021) — link
  52. Term Structure Models Driven by General Levy Processes (1999)
  53. Hot or Cold? A Comparison of Different Approaches to the Pricing of Weather Derivatives (2005)
  54. Estimation Methods of Reference Evapotranspiration at Unsampled Locations
  55. Option Pricing by Esscher Transforms (1994)
  56. Option Pricing in a Regime Switching Stochastic Volatility Model (2017)
  57. Pricing Rainfall Derivatives at the CME (2013)
  58. Levy processes in finance: Theory, numerics, and empirical facts (2000)
  59. Weather derivative valuation: the meteorological, statistical, financial and mathematical foundations (2005)
  60. Modeling of air temperatures: preprocessing and trends, reduced stationary process, extremes, simulation (2015)
  61. kdensity: Kernel Density Estimation with Parametric Starts and Asymmetric Kernels. R package version 1.0.1 (2019)
  62. Forecast based Pricing of Weather Derivatives (2012)
  63. Weather forecasting for weather derivatives (2005)
  64. A comparison of temperature density forecasts from GARCH and atmospheric models (2004)
  65. The use of weather forecasts in the pricing of weather derivatives (2003)
  66. Weather Derivatives (2013)
  67. Weather Derivatives Pricing: Modeling the Seasonal Residual Variance of an Ornstein-Uhlenbeck Temperature Process with Neural Networks (2009)
  68. On Modelling and Pricing Weather Derivatives (2010)
  69. A stochastic volatility model for the valuation of temperature derivatives (2023)
  70. Weather Forecasting for Weather Derivatives (2005) — link
  71. A comparison of temperature density forecasts from GARCH and atmospheric models (2004) — link
  72. Pricing Weather Derivatives in Cuba: A Proposal (2023)
  73. Pricing rainfall futures at the CME (2013)
  74. Neural Networks for Parameter Estimation in Intractable Models (2021)
  75. Pricing Temperature Derivatives under a Time-Changed Levy Model (2020)
  76. Time Series Analysis With Applications in R (2018)
  77. Hands-On Machine Learning with Scikit-Learn, Keras & TensorFlow Concepts, Tools, and Techniques to Build Intelligent Systems (2023)
  78. An Introduction to Statistical Learning with Applications in R (2013)
  79. Climate of chaos: Stanford researchers show why heat may make weather less predictable (2021) — link
  80. Accelerating Deep Network Training by Reducing Internal Covariate Shift (2015) — link
  81. A stochastic volatility model for the valuation of temperature derivatives (2023)
  82. Valuation of Precipitation Contracts (2021)
  83. The volatility of temperature and pricing of weather derivatives (2007)
  84. On Modelling and Pricing Rainfall Derivatives with Seasonality (2011)
  85. Dynamical Pricing of Weather Derivatives (2002)
  86. Generalized Autoregressive Conditional Heteroskedasticity (1986)
  87. R-vine Models for Spatial Time Series with an Application to Daily Mean Temperature (2015)
  88. Modeling and Forecasting Temperature in Toronto Using Time-Series and Machine-Learning Methods (2023)
  89. Machine Learning Methods for Weather Forecasting: A Survey (2024) — link
  90. Deep Learning and Foundation Models for Weather Prediction: A Survey (2025) — link
  91. Interpretable Machine Learning for Weather and Climate Prediction: A Survey (2024) — link
  92. Foundation Models for Weather and Climate Data Understanding: A Comprehensive Survey (2023) — link

0 of 100 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.