Kion QuantResearch
The principles behind our strategies, with the mathematics for those who want it.
Specific strategies are discussed privately with clients. What follows is the framework every one of them must satisfy.
01 · Independent return
Separating what the market gives from what selection adds.
Any portfolio's return divides into the part explained by the market, scaled by its sensitivity to it (beta), and the part left over (alpha). An index fund is almost entirely beta. Paying active fees for beta is the most common inefficiency in private portfolios.
return = α + β × market return + noise
02 · Breadth
Why many small edges are more dependable than a few large ones.
The fundamental law of active management holds that risk-adjusted skill grows with the square root of the number of independent decisions. A modest edge applied hundreds of times is statistically more reliable than a strong conviction applied twenty times.
information ratio ≈ skill per decision × √independent decisions
03 · Risk budget
A fixed loss per position keeps any losing streak survivable.
Every strategy experiences losing streaks. Their damage depends on how much each position is sized to lose, which we set before entry. The arithmetic is unforgiving the other way: a 50% loss requires a 100% gain to recover.
capital after n losses = (1 − risk per position)n
04 · Capacity
Trading cost rises with size, and fastest for the largest funds.
Research on institutional trading finds that market impact grows roughly with the square root of an order's share of daily volume. Doubling assets raises the cost of every trade by about 40%, and for the largest funds that cost eventually exceeds the edge being traded.
impact ≈ σ × √(order size ÷ daily volume)
Illustration: 10% of the portfolio in one stock trading $50 million a day, with 3% daily price variation.
05 · Validation
How we tell a durable edge from an overfit backtest.
Walk-forward testing. Each strategy is built on one window and evaluated on the following, unseen window, repeated across the full history.
A sealed recent period. The latest months are withheld from every design decision and used only to confirm.
Multiple-testing penalties. The more variations examined, the higher the bar each must clear, so the best of many random ideas is not mistaken for a real one.
Live monitoring. Every live result is compared with its expected range; persistent deviation leads to lower risk or retirement, never a quiet refit.