AI Training Data for Global Macro -- Signal Data

Global macro provides an excellent vertical for deploying AI because, by definition, it requires practitioners to evaluate multiple inter-related datasets across multiple countries. AI processes can evaluate disparate datasets and detect insights far faster than humans. The training process becomes significantly more operationally efficient and accurate if

If you already know which signals matter to you (which most global macro portfolio managers do), then including those signals as part of your training data delivers significant increases as well as more accurate outcomes from the beginning.

Global Macro Signal Data -- Overview

Global macro signal data, by definition, is higher value data because it incorporates an analytical layer. It also delivers operational efficiencies by accelerating the user's ability to transform data/information into actionable intelligence.

Global macro signal data can be challenging to configure in part because it requires a deep understanding of how reaction functions trickle through multiple disciplines (economics, geopolitics, domestic politics) simultaneously. Offloading the signal detection to AI can be helpful for some with limited background and experience in the field because AI can detect under-appreciated correlations and relationships faster.

But the real force multiplier is for subject matter experts who know with great precision the signals that matter for their specific use case. AI systems move these experts from hunter-gatherer mode so they can allocate their most valuable commodity (time) to higher value analytical endeavors.

Finally, signal data is always and everywhere derivative of (i) volume data and (ii) prior analytical work. Signal data provides users and their AI systems with a fast track to identifying when a tradeable inflection point has materialized. It is therefore higher up the value chain relative to volume data, but cannot exist without the volume data.

Signal Data -- Alerts and Event Detection

In order for an AI system to send an alert or detect an unusual event, it needs to evaluate historical data. The analytical process identifies repeatable patterns AND deviations from the pattern. Given enough raw data, AI systems can crunch the numbers quickly and identify possible inflection points. Dashboards and device-specific deployments provide classic mechanisms for delivering alerts and event detection, but generative AI can also be trained to provide verbal cues to users when specific activity of interest to the user has been detected.

Signal Data -- Trend Detection

Configuring trend detection signals requires AI systems at a minimum to use foundation data (e.g., volume, velocity) in a time series format to detect changes in activity levels. More sophisticated trend detection calculates correlation coefficients among multiple data sets.

Within language data, correlations can be detected across topics as well as other modifiers. For example, sentiment analysis relies on adjectives that express emotion or normative values in order to generate a sentiment score. Knowledge graphs and vector databases calculate correlation co-effients for topics and concepts. Generative AI uses those correlations to increase the accuracy of the output answers for human readers, but linguists can use the coefficients for more sophisticated analysis of the underlying language in question. We are just at teh front end of the innovation cycle for the kinds of analysis possible here when these tools are deployed against public policy language.

The bottom line is that the trend if your friend. If AI can help you spot the trend faster and more accurately, you have an informational advantage. This is true for advocates as well as for portfolio managers and FX traders, risk managers, investment analysts, and wealth managers...although the time horizon for trend detection can vary across use cases.

Signal Data -- Anomaly Detection

Analysis and Opinion

Anomaly detection technically is a form of trend detection, but it focuses on areas where at least two moving items diverge. When divergences occur on a repeatable, systemic basis, they are termed co-variances and can be used to inform investment strategies as much as correlations.

Stock market traders also rely on technical price movement analysis (e.g., Moving Average Convergence Divergence analysis) to identify when market shifts unrelated to a tradeable instrument's underlying/fundamental value create short-term profit opportunities. These kinds of market anomalies often are short-lived, rendering MACD analysis most relevant for systematic volatility traders and other highly specialized portfolio strategies.

Anomalies occur both in language and in quantitative contexts. One of the many advantages of using generative AI is that the underlying technology makes it easy to identify linguistic anomalies using mathematics. Traditional outputs involve responses designed to emphasize the most common or expected linguistic combinations. But separate prompts can be crafted to identify linguistic anomalies that have a high informational or signal content to the user.

The quantitative and ontological components of PolicyScope data accelerate and intensify anomaly detection with respect to public policy shifts at all frequency levels. BCMstrategy, Inc. works with its customers to optimize these kinds of configurations based on BCMstrategy, Inc.'s deep knowledge of public policy processes.

Anomaly detection need not occur on a high frequency or repeatable basis in order to be valuable. For low frequency events, anomaly detection should be paired with either alerts and signals or automated agentic AI processes in order to ensure that the informational value of anomaly has been detected and has informed decisions in a timely manner.

Signal Data -- Trend Projection

Welcome to crystal ball territory. AI's capacity to spot repeatable patterns makes it a highly effective and objective mechanism for delivering reliable estimates of likely future outcomes within a confidence interval. The superior computing power -- particularly parallel processing -- makes it easier to estimate reaction functions across a broad range of inter-related items simultaneously.

Real humans make decisions based on a wide range of factors that shift constantly. AI provides a mechanism that mimics that multifactor decision process and delivers outputs that reflect both the most common outcome and the outlier outcomes.

A conundrum exists. If trend projection occurs on an "unsupervised" basis, the rational for the trend projection may not be transparent, making it difficult for humans to evaluate the output. If trend projection occurs on a "supervised" basis, the machine is constrained to delivering outcomes that are consistent with past inputs.

Language processing which converts words into numerical integers and vectors (generative AI) and PolicyScope Data which converts public policy words into different kinds of numerical integers open up new possibilities for trend projection regarding public policy on a par with market reaction functions.

---

BCMstrategy, Inc. generates quantitative time series data and structured language data from public policy using a patented, award-winning process. Designed from the beginning to be used as ML/AI training data to support automated policy trend projection, the data is optimally structured to support deployment into automated research assistant applications powered by Generative AI. The company currently generates data within four thematic verticals: Trade Policy (TRD) |Monetary Policy (macroVS) | Climate/Energy Policy (CRRM3) | Digital Currency Policy (DCVS).