Machine learning
Attention models for forecasting digital-asset volatility
Transformer architectures were built for language, but the same attention mechanism turns out to be useful for weighting which past price windows matter to the next move. Where it helps, and where a much simpler model still wins.
Method9 min read
Regulation
The CSA crypto trading platform framework, in plain terms
What the Canadian Securities Administrators' pre-registration regime actually requires of platforms serving Canadian residents, what it does not cover, and the questions worth asking any venue that holds your assets.
Markets11 min read
Machine learning
Position sizing with reinforcement learning: what survived out of sample
An agent that learns how much to allocate, rather than what to buy. Promising in simulation, humbling in production — and the three constraints we had to add before it was safe to run on a client mandate.
Method13 min read
Macro
Bank of Canada policy and the CAD–Bitcoin correlation
The loonie is a commodity currency and Bitcoin is a risk asset, so overnight rate decisions reach both — but not through the same channel, and not on the same lag. Measuring the relationship across the last four tightening cycles.
Markets10 min read
Machine learning
Detecting quote stuffing without labelled data
Manipulation is rarely labelled, which makes it a clustering problem rather than a classification one. How unsupervised methods flag order-book behaviour that does not fit any legitimate trading pattern.
Method8 min read
Markets
Bitcoin after the halving: how second halves usually behave
Four halvings is not a sample size, and anyone presenting it as one is selling something. What the record does and does not support about the twelve months following a supply reduction.
Markets9 min read
Machine learning
Bilingual sentiment extraction from Canadian market coverage
Financial coverage in Canada runs in two languages, and a model trained only on English text misses half the signal in Quebec. Building a sentiment pipeline that treats French coverage as a first-class input.
Method12 min read
Infrastructure
Layer-2 settlement and what institutional adoption actually requires
Throughput was never the blocking constraint. Finality guarantees, bridge risk and the custody question are — and until those are answered, settlement volume stays where it is.
Markets10 min read
Method
When a model should be retired: our out-of-sample rules
Every strategy decays. The interesting question is how quickly you are willing to admit it. The specific thresholds that take a model out of client accounts, written down in advance so the decision is not made under pressure.
Method7 min read
Markets
Reading the S&P/TSX through an index CFD
Canada's benchmark is unusually concentrated in financials and energy, which changes what an index position is actually exposed to. What that means for anyone using it as a hedge rather than a directional bet.
Markets8 min read