Attention layers and the problem of regime change
Why transformer-style models spot a shift in market character earlier than the recurrent architectures they replaced.
Research
What our models are learning
Notes from the desk on applied machine learning, Canadian market structure, and the regulatory ground shifting under digital assets.
The CSA platform framework, three years on
What pre-registration undertakings actually changed for Canadians holding digital assets, and what is still unsettled.
Reading two languages of newsflow at once
Building a sentiment pipeline that treats French and English coverage as one signal rather than two disconnected feeds.
Bitcoin, the loonie and the correlation that keeps breaking
CAD-denominated exposure behaves differently from USD exposure, and the gap widens exactly when it matters most.
Discipline beats conviction in a falling market
Backtests across four drawdowns show where a rules-based system holds up and where a human reliably does not.
Spot crypto ETFs on the TSX and what they absorbed
Listed products pulled a specific kind of flow out of exchanges. Here is what that did to the order book.
Position sizing as a reinforcement learning problem
Treating exposure as the action space, rather than direction, changes what the model optimises for.
Bank of Canada policy and the risk-asset bid
Mapping overnight rate decisions against digital asset flows over eight consecutive announcement windows.
Finding wash trading without a labelled dataset
Unsupervised clustering surfaces manipulated volume days before the price break that follows it.
Pricing protocol risk in DeFi yields
An attractive rate is a claim about risk. Separating the technical component from the regulatory one is the hard part.