Stochastic Predictive Analytics for Stocks in the Newsvendor Problem

Pedro A. Pury · 2025

Read Stochastic Predictive Analytics for Stocks in the Newsvendor Problem by Pedro A. Pury (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-12397.txt.

Introduction

Pedro A. Pury builds a stochastic model for the classic newsvendor inventory problem, where a seller must stock goods before knowing demand. The model tracks how the distribution of stock levels changes over time without assuming any specific shape for demand. That flexibility matters when history is thin and forecasts must be short term.

Because no fixed demand law is imposed, the approach can fit varied real settings where a guessed distribution would mislead. Pury evaluates the model on real data from a large electronic marketplace, showing it works in a practical forecasting case. The work offers inventory managers a tool shaped for limited data and quick decisions.

Catalog record

CREATORPedro A. Pury
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE50 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-12397

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Stochastic Predictive Analytics for Stocks in the Newsvendor Problem Pedro A. Pury Abstract This work addresses a key challenge in inventory management by developing a stochastic model that describes the dynamic distribution of inventory stock over time without assuming a specific demand distribution. Our model provides a flexible and applicable solution for situations with limited historical data and short-term predictions, making it well-suited for the Newsvendor problem. We evaluate our model’s performance using real-world data from a large electronic marketplace, demonstrating its effectiveness in a practical forecasting scenario. Keywords: Stochastic data modeling, time series analysis, retail forecasting, Markov process MSC [2020] 90B05; 90B90; 60J10; 62R07 1 Introduction The Newsvendor problem is a fundamental model in inventory management ( Rossi, 2021 ) that accommodates both known ( Dvoretzky et al., 1952a ) and unknown ( Dvoretzky et al., 1952b ) demand distributions. Since its inception ( Edgewort, 1888 ) , it has been widely applied in inventory control and policy-making ( Arrow et al., 1951 ) , as well as various real-world situations ( Choi, 2012 ; Chen et al., 2016 ) . Its simplicity stems from considering a single product for sale, for which the optimal initial stock level must be determined to satisfy forecasted demand over a given period without restocking. The interplay among purchasing cost, selling price, and stock ordered at the beginning of the period determines the inventory management policies ( Whitin, 1952 ; Rosenblatt, 1954 ; Petruzzi and Dada, 1999 ) . The model has been extensively studied for single stock-keeping units (SKUs). Electronic marketplaces introduce an extra complication to the problem, as they need to manage a large number of SKUs at distribution centers alongside highly variable demand received through electronic platforms. This situation requires dealing with short demand histories to forecast in a short time horizon and designing forecasting methods that are easily upgradable with new data and have low computational costs for inference. Usually, forecasting methods in this context are intended for point forecasts of stock value. However, for practical and fundamental reasons, it is highly convenient to know the distribution of stock over time, given the initial level. This is a valuable tool for planning and decision-making in a wide variety of inventory situations. Without replenishment, the Newsvendor’s stock over time follows a non-increasing discrete-time series. Stochastically non-increasing Markov chains have been studied for a long time ( Daley, 1968 ) , and in recent years, particular attention has been given to scaling limits ( Miermont, 2011 ) . However, explicit solutions in specific cases, such as the Newsvendor problem, are currently lacking. This study contributes to the existing literature by developing a stochastic model that describes the dynamic distribution of inventory stock over time for a single stock-keeping unit (SKU) within a single-period planning horizon. Drawing inspiration from Croston’s method ( Croston, 1972 ) , which decomposes demand into sizes and intervals for forecasting, we construct the time distribution of stock by analyzing stochastic sales dynamics and demand distribution separately. This approach yields a time distribution that is agnostic to the demand distribution considered. We further innovate by deriving specific instances of the model for various parametric and non-parametric demand distributions, showcasing its flexibility and applicability. Our approach is well-suited for the Newsvendor problem, particularly in situations where historical data are limited and short-term predictions are required. By understanding the distribution of stock over time, inventory managers can make more informed decisions about inventory optimization, supply chain management, and resource allocation, ultimately leading to reduced costs and improved efficiency through better restocking policies. We evaluate our model’s performance in a real-world forecasting scenario using actual data from an electronic marketplace covering a wide range of SKUs. We demonstrate the effectiveness of our approach in the practical forecasting scenario of predicting stockouts, providing valuable insights for both academics and decision-making practitioners in retail management. Thus, the work is intended to present a useful forecasting case, but the generic nature of the methodology is emphasized. This paper is organized as follows: Sections 2 , 3 and 4 cover Materials and Methods. In Section 2 , we introduce our stochastic model, derive the solution for the time-dependent stock distributions, and analyze frustrated sales due to stockouts. In Section 3 , we complement the general stochastic framework by considering several demand distributions, ranging from a simple count of daily sales frequencies to various parametric demand distributions, including deterministic, Poisson, binomial, and negative binomial models. Section 4 applies these theoretical results to stockout forecasting, encompassing evaluation metrics, establishment of a baseline value, description of the dataset used in the experiments, and proposal of a benchmark for evaluations. Furthermore, we describe implementation methods for the naive frequentist non-parametric approach to demand modeling, as well as parameter estimation techniques for the distributions mentioned in the previous section. The results of stockout forecasting are presented in Section 5 , followed by conclusions in Section 6 . 2 Stochastic Model for Stocks In the absence of replacements or repositions, we can model the daily stock of a specific SKU as a monotonic non-increasing discrete-time series of non-negative integer values. The object of our interest is the probability, P(n,k) , of having in stock n items on day k , given an initial supply of m>1 items, given the initial condition P(n,k=0)=\delta_{n,m}\,, (1) where \delta_{n,m} is the Kronecker delta, and with the condition of normalization \sum_{n=0}^{m}P(n,k)=1,\;\forall k\geq 0. (2) The stochastic shopper’s demand is characterized by the probability per unit time, \alpha_{l} , which represents the probability of selling l units of the product within a day. In this case, the condition of normalization results \displaystyle\sum_{\ell=0}^{\infty}\alpha_{\ell}=1\,. (3) We further define the probability per unit time, \beta_{n}=\displaystyle\sum_{\ell=n}^{\infty}\alpha_{\ell} , of selling at least n items within a day. Specifically, \beta_{n} is associated with the shoppers’ intention to independently purchase at least n items on that day, irrespective of the current stock level. It follows from Eq. ( 3 ) that \beta_{0}=1 . 2.1 Markov model We assume that \alpha_{\ell} is independent of the current stock level. Furthermore, we assume that the shopper’s purchasing behavior remains consistent throughout the observation period, allowing us to consider as \alpha_{\ell} constant within this window. The assumption that \alpha_{\ell} remains constant over time is a simplification, as day-of-week seasonality may be present for some products. When dealing with the short length of demand series, more complex models may not be feasible, but the potential limitations of this approach should be recognized. Under these assumptions, we have a Markov process and we can write the following recurrence equations for the probabilities of interest \displaystyle P(0,k) \displaystyle= \displaystyle\sum_{n=0}^{m}\beta_{n}\,P(n,k-1)\,, (4) \displaystyle P(n,k) \displaystyle= \displaystyle\sum_{\ell=n}^{m}\alpha_{\ell-n}\,P(\ell,k-1),\mbox{ for }1\leq n\leq m\,. (5) In the Newsvendor context, P(0,k) represents the first-passage or absorption time distribution to the stock level 0 , which corresponds to the stockout time distribution. Using these recurrence equations, one can easily prove by complete induction the normalization given by Eq. ( 2 ), where the base case follows directly from Eq. ( 1 ). Alternatively, we can assume that the normalization condition is satisfied and prove by complete induction the recurrence Eq. ( 4 ) from Eq. ( 5 ), where the base case also follows directly from the initial condition.

Read the complete text