*Inflation, year-over-year percent change.
*The dataset and analytical data are refreshed instantly.
*This study leverages neural networks combined with game-theoretic reinforcement learning to nowcast the US Consumer Price Index (CPI) in real-time. By modeling inflation dynamics as a strategic game between economic agents and market forces, the algorithm processes high-frequency indicators to deliver faster, more adaptive, and highly accurate inflation forecasts ahead of official releases. Benchmarks
ML Model Testing : Game-Theoretic Reinforcement Learning (GTRL)
Hypothesis Testing : Shapley Additive exPlanations
Surveillance : US Macro
AUC Score :
Research Scope
The objective of this research is to develop a real-time machine-learning framework for predicting the U.S. Consumer Price Index (CPI) and, more broadly, identifying changes in inflationary conditions before the official CPI release. The study focuses on whether artificial neural networks can extract nonlinear relationships from a large set of economic, financial, market, and potentially sentiment-based indicators and use these relationships to generate timely inflation forecasts.The central motivation is the information delay inherent in conventional macroeconomic analysis. Official CPI statistics are released periodically, whereas many economic and financial variables are available at higher frequencies. A neural-network-based system can potentially combine these high-frequency signals with historical inflation data to estimate the current inflationary state before the official CPI figure becomes available.
Methodology
This study develops a real-time U.S. Consumer Price Index (CPI) prediction framework based on artificial neural networks (ANNs), reinforced by game-theoretic and reinforcement-learning techniques. The primary objective is to estimate current and forthcoming inflation dynamics using economic and financial information available before the official CPI release. The methodology is designed as a sequential prediction framework in which real-time macroeconomic signals are transformed into an inflation forecast, evaluated against realized CPI outcomes, and continuously improved through model optimization.To extend the forecasting framework toward a more behaviorally grounded representation of economic activity, individual artificial intelligence agents can be modeled as autonomous decision-making entities that learn to behave as heterogeneous economic individuals. Each agent is assigned a state vector representing its economic environment, including variables such as income, prices, inflation, interest rates, employment conditions, and other relevant macroeconomic indicators. Based on its observed state, the agent learns a consumption and expenditure policy through reinforcement learning. Rather than imposing a predetermined behavioral rule, the agent updates its decision policy according to the consequences of previous expenditure decisions, allowing heterogeneous patterns of consumption behavior to emerge endogenously. In this framework, agents may differ in their income levels, preferences, financial constraints, expectations, and sensitivity to macroeconomic conditions, thereby creating a population of economically heterogeneous decision-makers.
The aggregate response of the economy can then be obtained by combining the expenditure decisions of individual agents. At each period, changes in macroeconomic variables modify the state observed by the agents, leading them to adjust their consumption and spending decisions according to their learned policies. For example, an increase in inflation or interest rates may alter an agent's propensity to consume, while changes in income or employment conditions may affect its expenditure capacity. The aggregate expenditure generated by the population of agents can subsequently be linked to changes in macroeconomic outcomes, including consumption, demand, and ultimately inflation. This creates a feedback mechanism in which macroeconomic conditions influence individual spending behavior, while the aggregated decisions of individuals feed back into the macroeconomic environment. Such an agent-based reinforcement-learning architecture allows the model to capture the emergence of aggregate economic dynamics from the adaptive behavior of individual artificial agents rather than assuming that aggregate relationships are directly specified in advance.
The neural-network component is further integrated with a game-theoretic reinforcement-learning framework. In this setting, alternative predictive models or forecasting strategies can be treated as competing agents whose performance is evaluated according to their forecasting errors. The reinforcement-learning mechanism receives feedback from previous predictions and rewards models or strategies that produce more accurate forecasts. Consequently, the system can learn which predictive configuration performs best under changing economic conditions rather than relying exclusively on a fixed model specification. This approach is consistent with KappaSignal's broader methodology, which combines neural networks, game theory, reinforcement learning, model selection, and decision functions.
Game-theoretic methods are also used to improve the interpretability of the neural-network predictions. In particular, feature-contribution methods based on Shapley values can be used to estimate the marginal contribution of individual economic variables to a particular CPI prediction. This makes it possible to identify which variables are most strongly associated with an expected increase or decrease in inflation. For example, the model can determine whether energy prices, wages, housing costs, producer prices, or financial conditions made the largest contribution to a given inflation forecast. Therefore, the methodology is not limited to predicting the CPI level but can also provide information about the economic signals underlying the prediction. To evaluate predictive performance, the model is tested using out-of-sample and pseudo-real-time forecasting procedures. Performance can be assessed using metrics such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), directional accuracy, forecast bias, and, where probabilistic predictions are generated, calibration measures such as the Brier score. The neural-network model is also compared with conventional benchmark models, such as autoregressive and linear regression models, to determine whether the nonlinear machine-learning approach provides a statistically and economically meaningful improvement in inflation forecasting.
Overall, the methodology combines real-time economic data, nonlinear neural-network modeling, reinforcement learning, and game-theoretic analysis into a unified inflation nowcasting framework. The neural network provides the core mechanism for learning complex relationships between economic indicators and CPI, while reinforcement learning enables the forecasting system to adapt based on prediction outcomes and game-theoretic methods improve model selection and interpretability. The resulting framework therefore aims not only to predict the next CPI observation but also to identify emerging inflationary pressures before they are fully reflected in the official CPI release.
For further details on the methodology, neural network framework, reinforcement learning approach, and game-theoretic model selection, please refer to the KappaSignal Methodology documentation.
U.S. Consumer Price Index (CPI) Basket: Categories, Components & Weights
| CPI Category | Subcategory | Weight (%) |
|---|---|---|
| All Items | Total CPI | 100.000 |
| Food & Beverages | Food | 13.698 |
| Food at Home | 8.325 | |
| Food Away from Home | 5.373 | |
| Food at Home | Cereals and Bakery Products | 1.035 |
| Meats, Poultry, Fish & Eggs | 1.995 | |
| Dairy & Related Products | 0.758 | |
| Fruits & Vegetables | 1.269 | |
| Beverages | Nonalcoholic Beverages | 0.995 |
| Alcoholic Beverages | 0.840 | |
| Housing | Housing | 44.469 |
| Shelter | 35.625 | |
| Rent of Primary Residence | 7.840 | |
| Owners' Equivalent Rent | 26.204 | |
| Lodging Away from Home | 1.289 | |
| Tenants' & Household Insurance | 0.292 | |
| Fuels & Utilities | 4.546 | |
| Water, Sewer & Trash Collection | 1.144 | |
| Household Furnishings & Operations | Household Furnishings & Operations | 4.298 |
| Furniture & Bedding | 0.884 | |
| Appliances | 0.191 | |
| Household Equipment & Furnishings | 0.566 | |
| Tools & Outdoor Equipment | 0.691 | |
| Housekeeping Supplies | 0.847 | |
| Apparel | Apparel | 2.368 |
| Men's Apparel | 0.472 | |
| Women's Apparel | 0.787 | |
| Boys' Apparel | 0.116 | |
| Girls' Apparel | 0.141 | |
| Footwear | 0.585 | |
| Transportation | Transportation | 16.316 |
| Private Transportation | 14.832 | |
| New & Used Motor Vehicles | 7.204 | |
| New Vehicles | 3.838 | |
| Used Cars & Trucks | 2.759 | |
| Motor Fuel | 2.981 | |
| Gasoline | 2.895 | |
| Motor Vehicle Maintenance & Repair | 1.039 | |
| Motor Vehicle Insurance | 2.754 | |
| Medical Care | Medical Care | 8.423 |
| Medical Care Commodities | 1.489 | |
| Prescription Drugs | 0.973 | |
| Nonprescription Drugs | 0.381 | |
| Medical Care Services | 6.935 | |
| Physicians' Services | 1.684 | |
| Hospital & Related Services | 2.618 | |
| Health Insurance | 0.890 | |
| Recreation | Recreation | 5.137 |
| Video & Audio | 1.053 | |
| Pets, Pet Products & Services | 1.158 | |
| Sporting Goods | 0.526 | |
| Photography | 0.064 | |
| Other Recreational Goods | 0.380 | |
| Other Recreation Services | 1.841 | |
| Education & Communication | Education & Communication | 5.846 |
| Education | 2.602 | |
| College Tuition & Fees | 1.351 | |
| Elementary & High School Tuition | 0.408 | |
| Day Care & Preschool | 0.699 | |
| Communication | 3.244 | |
| Telephone Services | 1.466 | |
| Wireless Telephone Services | 1.340 | |
| Internet Services & Electronic Information | 0.923 | |
| Other Goods & Services | Other Goods & Services | 2.902 |
| Tobacco & Smoking Products | 0.445 | |
| Personal Care Products | 0.674 | |
| Personal Care Services | 0.676 | |
| Legal Services | 0.270 | |
| Funeral Expenses | 0.166 | |
| Financial Services | 0.236 |
Short-Term Economic Expectations (Neural Networks Perspective)
Based on recent macroeconomic data and simulations driven by neural networks agents, the projected market outlook for the near-term presents four core expectations:Sticky Inflation Trajectory: The neural-network agents expect near-term inflation to remain elevated, in the 3.6%–3.8% range. This would remain significantly above the central bank’s official target, suggesting that the agents anticipate persistent inflationary pressures through the fall.
Household Budget Strain: The neural-network agents expect household financial pressure to increase in the near term, driven by a widening gap between income and spending. The agents project nominal household income to grow by approximately 3.0%, while expecting household expenditures to increase by 4.9%, pointing to a potential squeeze on purchasing power.
Weakening Labor Market Confidence: The neural-network agents expect a gradual deterioration in near-term labor-market confidence. The agents project the probability of job loss to rise to 11.2%, while expecting the likelihood of an unemployed individual finding a job to decline to 40.8%. Together, these expectations point to a more cautious employment outlook and suggest a potential softening in household confidence.
Persistent High Rates & Tight Credit: The neural-network agents expect the Federal Reserve to maintain relatively elevated interest rates in the near term, as persistent core inflation may limit the pace of monetary easing. Accordingly, the agents anticipate that credit conditions could remain somewhat restrictive, with bank credit, mortgages, and consumer loans potentially remaining less readily available in the near term.