ML Price Prediction Model Development

Brent crude loses 15% in a month and recovers in a quarter—a forecasting error costs millions. A retail chain with a thousand SKUs loses up to 8% of revenue due to suboptimal pricing. ML price prediction models reduce MAPE to 3–5% on a one-month horizon. For example, in grain trading, a ±2% forecast accuracy saves up to $500,000 per large deal. In retail, properly tuned dynamic pricing increases margins by 7–15% without losing sales volume. We have built dozens of such models for commodities, real estate, and retail. Below is what actually works in production and how to deploy ML in your business. Want the same result? Request a data analysis—we will select the optimal model architecture.

Typology of Price Prediction Tasks

Commodities: Oil, gas, metals, agricultural products. Factors: futures curves, inventories, geopolitics, weather. Horizon: 1–12 months.

Real Estate: Object valuation (hedonic pricing model) or market index forecast. Horizon: 3–24 months.

Financial Assets: Stocks, currencies, cryptocurrencies. Most competitive environment—Efficient Market Hypothesis limits predictability. Practical horizon: 1–30 days.

Retail Prices: Optimal price prediction to maximize revenue/margin, considering price elasticity.

How We Build Price Prediction Models

Price Decomposition into Components

• Long-term trend (autocorrelation, macro factors)
• Seasonality (annual, quarterly, intra-week)
• Cyclicality (business cycles)
• Residual (irregular events, noise)

STL/MSTL decomposition is the first step in any price series analysis.

Hedonic Pricing for Real Estate

The most practical implementation is an appraisal model. Comparison of popular approaches:

Features for real estate:

• Object characteristics: area, rooms, floor, year built, material
• Location: distance to metro, schools, center; crime index; school district rating
• Infrastructure: proximity scores to shops, parks, hospitals
• Market conditions: comparable transactions from last 6–12 months

Accuracy of modern hedonic models: MAPE 5–12% for residential real estate in cities with good data.

Commodity Price Forecasting (Oil Example)

Features:
- Futures curve slope (contango/backwardation)
- EIA crude oil inventories (weekly)
- Baker Hughes rig count
- USD Index (DXY)
- Geopolitical risk index (NLP from news)
- Lagged price series: t-1, t-7, t-30, t-365

LightGBM with rolling cross-validation. MAPE 3–8% on a weekly horizon is acceptable. On a 3-month horizon, error rises to 15–25%.

Look-ahead bias problem: All features must be available at forecast time. Futures curve at time t is OK. Current week inventory data is not (published with delay).

Why Dynamic Pricing Outperforms Static Price Lists

Price Elasticity and Retail Pricing

Not forecasting future market price, but finding the optimal price to maximize revenue:

Elasticity model:

log(Demand) = α + β × log(Price) + γ × log(Competitor_Price) + δ × Promotions + ε

β — price elasticity coefficient. Typically ranges from -0.5 to -3.0 depending on category.

Dynamic pricing model: LightGBM predicts demand at different price points. Optimizer (scipy.optimize) finds price = argmax(Price × Demand(Price)). Constraints: min/max price, MSRP, brand price image.

Bayesian optimization allows exploring the price space without revenue loss during training. Based on our data, implementing dynamic pricing in a 500-store chain yields additional profit of 3 to 8 million rubles per month. Find out how ML can boost your margins—contact us for a consultation.

Model Development Steps

1. Data analysis and feature engineering—collect historical prices, external factors, clean data, analyze correlations.
2. Build baseline—simple model (mean, ARIMA) to assess lower accuracy bound.
3. Develop production model—LightGBM/XGBoost with hyperparameter optimization, rolling cross-validation.
4. Test on holdout set—check for look-ahead bias, metric stability.
5. Package as REST API—endpoint for on-demand forecasts, Swagger documentation.
6. Monitor and A/B test—track MAPE, automatic retraining on degradation.

Comparison of Time Series Forecasting Methods

params:
  objective: regression
  metric: mape
  boosting_type: gbdt
  num_leaves: 127
  learning_rate: 0.05
  feature_fraction: 0.8
  bagging_fraction: 0.8
  bagging_freq: 5
  num_rounds: 1000
  early_stopping_rounds: 50

What's Included

• Data analysis and feature engineering
• Baseline and production model development
• REST API for integration into your system
• Documentation and team training
• Monitoring and post-deployment support

Integration and Updates

• Retail: Integration with pricing engine (Revionics, Wiser, custom) via API
• Commodities: Export forecasts to ERP/Treasury system for hedging
• Real Estate: REST API for AVM (Automated Valuation Model), integration with agency CRM

Monitoring: Track MAPE on a 30-day rolling window. On degradation >30%—automatic retraining.

Timelines: Hedonic real estate model or commodity price forecast—4–6 weeks. Retail dynamic pricing with price elasticity and A/B testing—3–4 months.

Our Experience and Guarantees

Our engineers have over 5 years of ML model experience, with more than 30 projects in forecasting. We guarantee metric transparency, experiment reproducibility, and full documentation. Our clients see a 5–15% margin increase after dynamic pricing implementation. Reducing MAPE from 20% to 5% saves millions in commodity trading.

The hedonic pricing approach is described on Wikipedia—an appraisal model.

Book a consultation—we will analyze your data and propose the optimal model for your business. Contact us to discuss your project.

When does a time series forecasting model fail in production?

The CFO requests a quarterly sales forecast. An analyst builds SARIMA on three years of data, achieves MAPE 8.3% on the test set, and deploys. Two months later, the metric in production jumps to 23%. The root cause: the model was trained on pre‑COVID data, tested on a stable period, but production hit a promotion and supply chain disruption. Data leakage plus distribution shift—perfect notebook numbers, a broken forecast in reality. We have seen this pattern dozens of times across retail, fintech, and IoT. Our team has delivered more than 50 forecasting projects over 5+ years.

Incorrect cross-validation. Standard train_test_split for time series creates data leakage: the model sees future values during training. The correct approach is TimeSeriesSplit or walk‑forward validation with an expanding window.

Multiple seasonality. Hourly electricity consumption has three seasonalities: daily (24h), weekly (168h), yearly (8760h). SARIMA handles only one. Prophet can handle multiple but scales poorly to thousands of series.

Missing values and anomalies. A missing sensor reading is information (the sensor turned off), not NaN. Linear interpolation destroys this signal. Proper handling depends on the missingness mechanism.

Cold start. A new SKU in a 50,000‑item assortment has no history, yet a forecast is needed. Standard approaches fail; cross‑learning or feature‑based methods are required.

Why is model selection critical for your data?

Prophet (Meta) – a solid start for business data with clear seasonality and holidays. Fast setup, interpretable, built‑in outlier detection. Fails on irregular patterns and does not scale beyond ~10k series without parallelization.

Gradient boosting on features (LightGBM, XGBoost) – often underestimated. Engineer lags (t‑1, t‑7, t‑28), rolling means, day‑of‑week, holidays. The model trains on all series simultaneously, solving cold start via transfer learning. MAPE in retail often beats neural nets with proper feature engineering.

TFT (Temporal Fusion Transformer) – a transformer designed for interpretable forecasting with covariates. Built‑in variable selection, temporal attention, quantile outputs. Available in pytorch‑forecasting. Requires ~10,000+ records per series for stable training.

PatchTST – splits the series into patches (like ViT for images), capturing local patterns better than classic transformers. Excellent for long‑horizon forecasting (96–720 steps ahead).

N‑HiTS, N‑BEATS – attention‑free neural architectures, faster than TFT, competitive accuracy. N‑BEATS won the M4/M5 benchmarks for tasks without covariates.

How do we deploy TFT in production?

A typical pipeline via pytorch‑forecasting:

training = TimeSeriesDataSet(
    data,
    time_idx="time_idx",
    target="sales",
    group_ids=["store", "sku"],
    min_encoder_length=max_encoder_length // 2,
    max_encoder_length=max_encoder_length,  # 120 days
    min_prediction_length=1,
    max_prediction_length=max_prediction_length,  # 28 days
    static_categoricals=["store_type", "category"],
    time_varying_known_reals=["price", "promo_flag"],
    time_varying_unknown_reals=["sales"],
    target_normalizer=GroupNormalizer(groups=["store", "sku"], transformation="softplus"),
)

A common mistake: the default target_normalizer (StandardScaler) breaks predictions for series with zero values (no sales on weekends). GroupNormalizer with transformation="softplus" is the correct choice for count data.

Case study: retail demand forecasting

A chain of 120 stores, 8,000 SKUs, 28‑day forecast horizon. The original system: SARIMA per series, MAPE 18.4%, retraining cycle – 6 hours. We replaced it with TFT on PyTorch + pytorch‑forecasting: a single model for all series, MAPE 11.2%, retraining – 40 minutes on an A10G. Feature importance via variable selection revealed that day_before_holiday influences more than the holiday date itself. Annual savings on inference alone exceeded $50,000.

Step‑by‑step configuration

1. Data collection and preparation. Handle missing values (mark NaN, interpolate only for technical failures), aggregate to required frequency, engineer covariates (holidays, promotions, prices).
2. Create TimeSeriesDataSet. Set group_ids (store + SKU), time index, forecast horizon. Choose target_normalizer based on target distribution.
3. Train a baseline. Prophet or LightGBM first – to understand complexity.
4. Train TFT. Use TemporalFusionTransformer with loss=QuantileLoss(), tune learning rate and hidden layer sizes.
5. Validate and interpret. Walk‑forward test, analyze variable selection, build attention heatmaps.

How to properly evaluate forecast quality?

RMSE alone is misleading – it over‑penalizes large values. Our standard set:

• MAPE – interpretable, unstable near zero.
• sMAPE – symmetric, avoids division by small numbers.
• MASE (Mean Absolute Scaled Error) – normalized relative to a naive seasonal forecast, ideal for comparing series of different scales.
• Pinball loss – for probabilistic forecasting, inventory management.

We guarantee a model card with these metrics on the validation set and walk‑forward results on at least 6 months of history.

What deliverables do you receive?

• Documentation of chosen architecture and hyperparameter rationale.
• Reproducible training and inference pipeline (Docker + CI/CD + Airflow/Prefect).
• Committed code with unit tests for key components.
• Team training: retraining, output interpretation, deployment of new versions.
• 3 months of post‑delivery support (consultations, bug fixes, fine‑tuning).

The model is deployed via FastAPI or Triton Inference Server. Retraining is scheduled (e.g., weekly) via Airflow with drift validation and automatic rollback if metrics deteriorate.

Process and timeline

We start with EDA: visualization, ADF test, STL decomposition, analysis of missing values and outliers. This takes 2–3 days but often reveals systemic data issues that block forecasting. Then we build a baseline (naive seasonal, Prophet), engineer features for LightGBM, and select a neural architecture if needed. Walk‑forward validation with a realistic horizon. Deployment via API with automatic retraining scheduled via Airflow or Prefect.

Timeline: MVP forecast on one data type – 3–6 weeks. Hierarchical forecasting system with automation – 2–5 months. Cost is calculated individually based on data volume, number of series, and required accuracy.

Our team consists of certified ML engineers (AWS ML Specialty, GCP Professional ML Engineer) with 5+ years on the market and over 50 completed forecasting projects. Contact us for a free analysis of your data – we will assess the task and provide initial recommendations within 1–2 days. Request a consultation to ensure your forecasts work in production, not just in a notebook.