Abstract: This paper first proposed a new model to describe the
relationship between two paired asset prices:
$a_x(t)X_t-a_y(t)Y_t=m_t+s_t\varepsilon_t,$ where $X_t$ and $Y_t$
denote the prices of two paired financial assets at time $t$,
$a_x(t)$ and $a_y(t)$ the matching coefficients, $m_t$ the long-term
trend, $s_t$ the standard deviation of residual, and $\varepsilon_t$
the standardized residual. When $a_x(t)$, $a_y(t)$, $m_t$ and $s_t$
are constants, the model is reduced to a kind of two-variable
cointegration model. Based on this new model, the paper proposed a
statistical arbitrage method for high-frequency trading using the
stationary process $\{\varepsilon_t\}$. As its application, this
method was used on three major ETFs in China financial markets and
achieved very stable and high revenue on all three pairs.

Key words: high frequency trading, statistical arbitrage, stationary process