Predicting Bond Prices: Regime Changes
summary
Real-time engine BondDroid TM predicts bid and ask prices for thousands of corporate bonds using an ensemble of roughly 500 experts with varied look-back windows, blending models by recent accuracy and retraining frequently on cloud instances. Models include regularized linear fits and nonlinear random forests, feature construction borrows pricing from companion bonds for illiquid issues, and backtests show ~78.5% of predictions within 5 basis points in low volatility and ~76.5% in higher volatility.
AI-written summary · about
chapters (8)
- 0:00Introduction to concept drift
- 0:35Challenges in market data
- 2:59Strategies for non-stationary
- 5:35The corporate bond market
- 9:45Machine learning architecture
- 13:05Minimizing cumulative regret
- 16:12Case study: bond evaluation
- 18:24Results and Q&A
related talks
30mPredictive trading
33mPredicting Cryptocurrency Exchange Rates
38mThe Journey Through Real Estate Price Model Development
33mUntangling the complexity of demand forecasting models: building a Market Simulator
35mCan Memes Predict Markets? Using Social Media Sentiment for Stock Forecasting
39mFuture of Forecasting the Future in...
08mQ n A with Dr Arun Verma
26mWin More Effectively in Real-Time Bidding