Multimodal Deep Learning for Merger Predictions

We are working on an academic research project in finance and machine learning. Our dataset includes:

Structured data from Compustat (financial statements)

M&A activity records from 1995 to 2019

A constructed supply network (based on firm relationships)

We aim to predict merger waves using a multimodal deep learning model.

Scope of Work:
Build an MLP (Multilayer Perceptron) that takes in financial statement variables.

Design a Graph Neural Network (GNN) using the supply chain network data.

Fuse the MLP and GNN embeddings (via concatenation or attention-based fusion).

Output a probability score of a firm participating in a merger wave.

Provide clean, well-commented code (preferably in PyTorch or TensorFlow).

Skills Required:
Experience with deep learning, MLP, and GNNs

Familiarity with financial or tabular datasets

Multimodal model fusion techniques

Knowledge of mergers & acquisitions or economic modeling (bonus)

Deliverables:
Source code for MLP, GNN, and fused model

Model training script with validation results

Documentation (README)

Optional: brief explanation of modeling choices