fynance.features.metrics.roll_sharpe¶
-
fynance.features.metrics.
roll_sharpe
(X, rf=0, period=252, w=None, log=False, axis=0, dtype=None, ddof=0)¶ Compute rolling sharpe ratio of each X’ series.
Parameters: - X : np.ndarray[dtype, ndim=1 or 2]
Time-series of prices, performances or index.
- rf : float or np.ndarray[dtype, ndim=1 or 2], optional
Means the annualized risk-free rate, default is 0. If an array is passed, it must be of the same shape than
X
.- period : int, optional
Number of period per year, default is 252 (trading days).
- w : int, optional
Size of the lagged window of the rolling function, must be positive. If
w is None
orw=0
, thenw=X.shape[axis]
. Default is None.- log : bool, optional
If true compute sharpe with the formula for log-returns, default is False.
- axis : {0, 1}, optional
Axis along wich the computation is done. Default is 0.
- dtype : np.dtype, optional
The type of the output array. If dtype is not given, infer the data type from X input.
- ddof : int, optional
Means Delta Degrees of Freedom, the divisor used in calculations is
T - ddof
, whereT
represents the number of elements in time axis. Default is 0.
Returns: - np.ndarray[dtype, ndim=1 or 2]
Serires of rolling Sharpe ratio.
See also
Notes
Sharpe ratio [7] is computed as the rolling annualized expected returns (
roll_annual_return
) minus the risk-free rate (noted \(rf\)) over the rolling annualized volatility of returns (roll_annual_volatility
) such that \(\forall t \in [1:T]\):\[sharpeRatio_t = \frac{E(R | R_{1:t}) - rf_t}{\sqrt{period \times Var(R | R_{1:t})}}\]Where, \(R_1 = 0\) and \(R_{2:T} = \begin{cases}ln(\frac{X_{2:T}} {X_{1:T-1}}) \text{, if log=True}\\ \frac{X_{2:T}}{X_{1:T-1}} - 1 \text{, otherwise} \\ \end{cases}\).
References
[7] https://en.wikipedia.org/wiki/Sharpe_ratio Examples
Assume a monthly series of prices:
>>> X = np.array([70, 100, 80, 120, 160, 80]).astype(np.float64) >>> roll_sharpe(X, period=12) array([ 0. , 10.10344078, 0.77721579, 3.99243019, 6.754557 , 0.24475518])