Precision forecasting is the key to maximizing revenues and optimizing collateral in the ever-evolving world of global securities finance. Advanced Securities Consulting (ASC) reached a milestone in predictive data analytics when the firm announced the spectacular results of Phase I for its 2024 Deep Learning Proof-of-Concept (POC I) at the IMN Global Securities Finance Conference on February 4th in Ft. Lauderdale, Florida.
For the first time, as reported by ASC, securities lenders were able to review compelling evidence that deep learning (DL) transformer models could use real-world data to outperform traditional AutoRegressive Integrated Moving Average (ARIMA) models in forecasting daily average rebates for cash-collateralized equity loans. Beyond simply proving technological superiority for DL models, the ASC POC I also validated the SEC’s controversial 10c-1 disclosure rule, demonstrating that forecasting models can rely solely on publicly available aggregates — specifically, those data fields that will be included in the Financial Industry Regulatory Authority’s (FINRA) upcoming SLATE disclosures.
This convergence of technological innovation and regulatory compliance is a significant step toward fulfilling the transparency mandates initially set out for U.S. “shadow banking” by the 2010 Dodd-Frank Act. Quite significantly, today’s data vendors, who already have access to even more granular loan-level information, will also benefit in that they should be able to refine their own forecasting models even further with tomorrow’s public data releases.
Deep Learning Models: A Superior Alternative to ARIMA in Securities Finance
ARIMA models have long been a staple in time-series forecasting. However, ARIMA struggles with complex, nonlinear relationships in dynamic financial environments, particularly in predicting securities lending rates. The ASC POC I highlights the clear advantages of deep learning transformer models, which:
- Capture Market Dynamics More Accurately – Unlike ARIMA, which assumes a linear dependence on past values, DL models dynamically adjust to changes in market conditions, capturing nonlinear interactions in targeted fees, rebates and yield spreads.
- Process High-Dimensional Data Efficiently – Transformer DL models were used to incorporate multiple interdependent variables in a weighted dataset of metrics for borrower demand, loan volume, and collateral types. ASC reported that the platform was designed to help human experts with AI assistants to manage in a more holistic ecosystem for predictive analytics.
- Optimize Fees and Rebates with the Goldilocks Rate Model™ – The ASC POC I advanced the concept of the “Goldilocks Rate,” an equilibrium fee/rebate rate that is positioned within the rate dispersion to maximize lender profitability without discouraging borrower demand. This approach to pricing elasticity is said to ensure a balanced cash and collateral management strategy that ARIMA models, with their rigidity, cannot achieve.
Regulatory Alignment: How ASC’s POC I Supports the SEC’s 10c-1 Rule
The SEC’s 10c-1a rule mandates greater transparency in securities lending markets, requiring lenders to disclose critical data elements related to loan pricing, volume, and counterparty risk. Historically, securities lending has been characterized by opacity, allowing intermediaries to extract hidden margins and maintain market inefficiencies. The ASC POC I proves that accurate and actionable forecasting can be achieved using only the variables that will be publicly available via FINRA’s SLATE.
This validation has two major implications:
- Evaluating Trading Efficiency Without Proprietary Data – ASC’s deep learning models rely only on data points that will be included in SLATE, such as loan volume, weighted-average rebates, and collateral classifications. By doing so, ASC demonstrates that market participants will be able to make precise rate forecasts without requiring access to proprietary lending data.
- Aligning with Dodd-Frank’s Intent to Reduce Opacity – One of the original goals of the Dodd-Frank Act was to enhance financial transparency and reduce systemic risks associated with opaque securities lending practices. By proving that publicly available data will be sufficient for accurate forecasting of rates and counterparty risks, ASC’s POC I aligns directly with this regulatory intent.
What This Means for Data Vendors
Like FINRA, the major data vendors will have access to full loan-level disclosures at the outset. This data will include confidential legal entity identifiers (LEIs) and certain transaction-level details that are not intended to be part of the public SLATE dataset. ASC’s POC I relied only on aggregated data from the FIS Lending Pit™ service, yet still demonstrated superior forecasting performance. With more granular loan data, vendors should be able to improve accuracy even further within the closed system of U.S. securities finance.
Vendors will be able to recalibrate their samples to match the full SLATE market census. However, existing methodologies may be complex to update for continuity and may not account for previously unreported transaction fields in the vendors’ existing give-to-get business models. Vendors can minimize selection biases and refine their rate predictions by incorporating a broader data set. Superior forecasts will become a key competitive differentiator.
Feature engineering and model selection will define the future of securities finance data analytics. Vendors with the best models will lead the market. Those market participants using data vendors who fail to adapt will struggle to compete on the basis of metrics. The introduction of SLATE data provides an opportunity for vendors to set new and better industry benchmarks.
What This Means for the Industry
The implications of the ASC POC I are far-reaching:
- For Lenders: Adopting DL-based forecasting models can improve profitability by enabling more accurate rate setting and collateral optimization, reducing reliance on existing borrower-favorable pricing.
- For Borrowers: Greater transparency in pricing data will enable borrowers to compare rates more effectively, fostering a more competitive and equitable marketplace.
- For Regulators: The ASC POC I provides empirical support for the SEC’s 10c-1 disclosure rule that mandates SLATE, reinforcing the idea that increased transparency can lead to more efficient pricing without creating unintended market distortions.
Conclusion: The Future is Now for Predictive Data Analytics in Securities Finance
ASC’s POC I is more than just a technological proof-of-concept — it is a validation of the regulatory push for greater transparency. By demonstrating that DL models significantly outperform ARIMA models and that these models can function using only public data, ASC is proving that the future of securities lending is one of precision, fairness, and regulatory alignment.
As the SLATE disclosures become standard, market participants who adopt deep learning models over the next several months will gain a competitive edge, leveraging AI-driven forecasts to optimize lending rates, collateral, and cash flows. The industry is moving beyond benchmarking toward predictive analytics — and ASC appears to have taken the lead. Data vendors must keep pace or risk falling behind.
As ASC refines these predictive capabilities, the next step is to establish a secure, real-time data trust that will revolutionize counterparty risk management and optimize collateral in a T+0 settlement environment.
Forthcoming Part 2: Data trusts will empower Gen AI use cases.