Insights
Market regime analysis & strategy deep dives
We write essays about how quantitative strategies actually work in real markets — including when and why they fail. New articles weekly.
16 strategies × 3 years of real data. Here's who won, who lost, and what that means.
We just finished a grid-optimized backtest on all 16 strategies against 3 years of OANDA daily bars. The Sharpe rankings will surprise you. Markov Switching came first (1.40) — above XGBoost. Pre-FOMC drift, the anomaly we added last month on the strength of a Journal of Finance paper, came dead last (-1.28 Sharpe). Here's the full scoreboard and what it tells us.
By Li Tan
Read articleThe /signals page went from 20 seconds to 500ms — here's the one change that did it.
We had a 14-strategy dashboard that took up to 30 seconds to load on a cold visit. Every click triggered 24 OANDA API calls, three MSR fits, an XGBoost train, and a Johansen test. Then we stopped doing any of that on request.
By Li Tan
Read articleWe gave carry trade real history. It had opinions.
Our carry strategy MVP used a single 2026 rate snapshot — JPY 0.73, USD 3.64 — as if that table had always been true. Connecting FRED's historical series changed the story: our signal count nearly 6x'd, and some long-term positions we thought were correct turned out to have been short the same currency a year earlier.
By Li Tan
Read articleMSR walk-forward? It depends.
We added an annual-refit mode to our Markov Switching strategy expecting it to dominate the 'fit once' version. It doesn't. On SPX it wins huge (+28% vs +1%). On gold it loses (+3% vs +33%). Engineering intuition betrayed us — here's what the 8-year data actually shows.
By Li Tan
Read articleOur XGBoost Sharpe just hit 2.14. We're suspicious.
Since January 1, XGBoost has posted +5.5% with 7 trades and a Sharpe north of 2. That's better than the 3-year optimized backtest. When a model performs above its own ceiling, the first question should be 'what's broken?'
By Li Tan
Read article71 days in. Four of our six strategies are flat or red.
The live track record has been running since January 1. Here's where we are, what's working, what isn't, and what I'd change (nothing).
By Yuko Xiu, CFA
Read articleWhy mean-reversion strategies are sleeping in 2026
Pair trading, cointegration baskets, and HMM regime models have all underperformed over the past 12 months. Here's why — and why that's exactly what you'd expect.
By Yuko Xiu, CFA
Read articleThe cointegration basket with zero trades (and why I'm not touching it)
One of our six strategies has executed 0 trades in 71 days. Here's what that actually means — and why it's the most honest result in the whole track record.
By Li Tan
Read articleWhat cointegration actually tests (and what most people get wrong)
Most pair trading tutorials hand-wave around the cointegration test. We walk through what it really checks, why correlation alone isn't enough, and how to spot a 'spurious' cointegration in real data.
By Li Tan
Read articleWhy TSMOM uses a 12-month lookback (and not 6 or 3)
The original Moskowitz, Ooi, Pedersen paper landed on a 12-month signal with a 1-month skip. We dig into the empirical reasons, the regime dependence, and what happens if you change the parameter.
By Yuko Xiu, CFA
Read articleWant new articles delivered? Upgrade to Learner to subscribe to the weekly market regime newsletter. See pricing →