import numpy as np from scipy.stats import norm, t from cryptotrader.utils import safe_div def fit_normal(ret): mu_norm, sig_norm = norm.fit(ret) return mu_norm, sig_norm def fit_t(ret): parm = t.fit(ret) nu, mu_t, sig_t = parm nu = np.round(nu) return mu_t, sig_t, nu def polar_returns(ret, k): """ Calculate polar return :param obs: pandas DataFrame :return: return radius, return angles """ ret= np.mat(ret) # Find the radius and the angle decomposition on price relative vectors radius = np.linalg.norm(ret, ord=1, axis=1) angle = np.divide(ret, np.mat(radius).T) # Select the 'window' greater values on the observation index = np.argpartition(radius, -(int(ret.shape[0] * k) + 1))[-(int(ret.shape[0] * k) + 1):] index = index[np.argsort(radius[index])] # Return the radius and the angle for extreme found values return radius[index][::-1], angle[index][::-1] # Pareto Extreme Risk Index def ERI(R, Z, w): # alpha alpha = safe_div((R.shape[0] - 1), np.log(safe_div(R[:-1], R[-1])).sum()) # gamma gamma = (1 / (Z.shape[0] - 1)) * np.power(np.clip(w * Z[:-1].T, 0.0, np.inf), alpha).sum() return gamma # Normal CVaR def CVaR(mu, sig, alpha=0.01): return alpha ** -1 * norm.pdf(norm.ppf(alpha)) * sig - mu # Student T CVaR def TCVaR(mu, sig, nu, h=1, alpha=0.01): xanu = t.ppf(alpha, nu) return -1 / alpha * (1 - nu) ** (-1) * (nu - 2 + xanu ** 2) * t.pdf(xanu, nu) * sig - h * mu """ The following risk and adjusted returns metrics belongs to: http://www.turingfinance.com/computational-investing-with-python-week-one/ """ import math import numpy import numpy.random as nrand """ Note - for some of the metrics the absolute value is returns. This is because if the risk (loss) is higher we want to discount the expected excess return from the portfolio by a higher amount. Therefore risk should be positive. """ def vol(returns): # Return the standard deviation of returns return numpy.std(returns) def beta(returns, market): # Create a matrix of [returns, market] m = numpy.matrix([returns, market]) # Return the covariance of m divided by the standard deviation of the market returns return numpy.cov(m)[0][1] / numpy.std(market) def lpm(returns, threshold, order): # This method returns a lower partial moment of the returns # Create an array he same length as returns containing the minimum return threshold threshold_array = numpy.empty(len(returns)) threshold_array.fill(threshold) # Calculate the difference between the threshold and the returns diff = threshold_array - returns # Set the minimum of each to 0 diff = diff.clip(min=0) # Return the sum of the different to the power of order return numpy.sum(diff ** order) / len(returns) def hpm(returns, threshold, order): # This method returns a higher partial moment of the returns # Create an array he same length as returns containing the minimum return threshold threshold_array = numpy.empty(len(returns)) threshold_array.fill(threshold) # Calculate the difference between the returns and the threshold diff = returns - threshold_array # Set the minimum of each to 0 diff = diff.clip(min=0) # Return the sum of the different to the power of order return numpy.sum(diff ** order) / len(returns) def var(returns, alpha): # This method calculates the historical simulation var of the returns sorted_returns = numpy.sort(returns) # Calculate the index associated with alpha index = int(alpha * len(sorted_returns)) # VaR should be positive return abs(sorted_returns[index]) def cvar(returns, alpha): # This method calculates the condition VaR of the returns sorted_returns = numpy.sort(returns) # Calculate the index associated with alpha index = int(alpha * len(sorted_returns)) # Calculate the total VaR beyond alpha sum_var = sorted_returns[0] for i in range(1, index): sum_var += sorted_returns[i] # Return the average VaR # CVaR should be positive return abs(sum_var / index) def prices(returns, base): # Converts returns into prices s = [base] for i in range(len(returns)): s.append(base * (1 + returns[i])) return numpy.array(s) def dd(returns, tau): # Returns the draw-down given time period tau values = prices(returns, 100) pos = len(values) - 1 pre = pos - tau drawdown = float('+inf') # Find the maximum drawdown given tau while pre >= 0: dd_i = (values[pos] / values[pre]) - 1 if dd_i < drawdown: drawdown = dd_i pos, pre = pos - 1, pre - 1 # Drawdown should be positive return abs(drawdown) def max_dd(returns): # Returns the maximum draw-down for any tau in (0, T) where T is the length of the return series max_drawdown = float('-inf') for i in range(0, len(returns)): drawdown_i = dd(returns, i) if drawdown_i > max_drawdown: max_drawdown = drawdown_i # Max draw-down should be positive return abs(max_drawdown) def average_dd(returns, periods): # Returns the average maximum drawdown over n periods drawdowns = [] for i in range(0, len(returns)): drawdown_i = dd(returns, i) drawdowns.append(drawdown_i) drawdowns = sorted(drawdowns) total_dd = abs(drawdowns[0]) for i in range(1, periods): total_dd += abs(drawdowns[i]) return total_dd / periods def average_dd_squared(returns, periods): # Returns the average maximum drawdown squared over n periods drawdowns = [] for i in range(0, len(returns)): drawdown_i = math.pow(dd(returns, i), 2.0) drawdowns.append(drawdown_i) drawdowns = sorted(drawdowns) total_dd = abs(drawdowns[0]) for i in range(1, periods): total_dd += abs(drawdowns[i]) return total_dd / periods def treynor_ratio(er, returns, market, rf): return (er - rf) / beta(returns, market) def sharpe_ratio(er, returns, rf): return (er - rf) / vol(returns) def information_ratio(returns, benchmark): diff = returns - benchmark return numpy.mean(diff) / vol(diff) def modigliani_ratio(er, returns, benchmark, rf): np_rf = numpy.empty(len(returns)) np_rf.fill(rf) rdiff = returns - np_rf bdiff = benchmark - np_rf return (er - rf) * (vol(rdiff) / vol(bdiff)) + rf def excess_var(er, returns, rf, alpha): return (er - rf) / var(returns, alpha) def conditional_sharpe_ratio(er, returns, rf, alpha): return (er - rf) / cvar(returns, alpha) def omega_ratio(er, returns, rf, target=0): return (er - rf) / lpm(returns, target, 1) def sortino_ratio(er, returns, rf, target=0): return (er - rf) / math.sqrt(lpm(returns, target, 2)) def kappa_three_ratio(er, returns, rf, target=0): return (er - rf) / math.pow(lpm(returns, target, 3), float(1/3)) def gain_loss_ratio(returns, target=0): return hpm(returns, target, 1) / lpm(returns, target, 1) def upside_potential_ratio(returns, target=0): return hpm(returns, target, 1) / math.sqrt(lpm(returns, target, 2)) def calmar_ratio(er, returns, rf): return (er - rf) / max_dd(returns) def sterling_ration(er, returns, rf, periods): return (er - rf) / average_dd(returns, periods) def burke_ratio(er, returns, rf, periods): return (er - rf) / math.sqrt(average_dd_squared(returns, periods)) # tests def test_risk_metrics(): # This is just a testing method r = nrand.uniform(-1, 1, 50) m = nrand.uniform(-1, 1, 50) print("vol =", vol(r)) print("beta =", beta(r, m)) print("hpm(0.0)_1 =", hpm(r, 0.0, 1)) print("lpm(0.0)_1 =", lpm(r, 0.0, 1)) print("VaR(0.05) =", var(r, 0.05)) print("CVaR(0.05) =", cvar(r, 0.05)) print("Drawdown(5) =", dd(r, 5)) print("Max Drawdown =", max_dd(r)) def test_risk_adjusted_metrics(): # Returns from the portfolio (r) and market (m) r = nrand.uniform(-1, 1, 50) m = nrand.uniform(-1, 1, 50) # Expected return e = numpy.mean(r) # Risk free rate f = 0.06 # Risk-adjusted return based on Volatility print("Treynor Ratio =", treynor_ratio(e, r, m, f)) print("Sharpe Ratio =", sharpe_ratio(e, r, f)) print("Information Ratio =", information_ratio(r, m)) # Risk-adjusted return based on Value at Risk print("Excess VaR =", excess_var(e, r, f, 0.05)) print("Conditional Sharpe Ratio =", conditional_sharpe_ratio(e, r, f, 0.05)) # Risk-adjusted return based on Lower Partial Moments print("Omega Ratio =", omega_ratio(e, r, f)) print("Sortino Ratio =", sortino_ratio(e, r, f)) print("Kappa 3 Ratio =", kappa_three_ratio(e, r, f)) print("Gain Loss Ratio =", gain_loss_ratio(r)) print("Upside Potential Ratio =", upside_potential_ratio(r)) # Risk-adjusted return based on Drawdown risk print("Calmar Ratio =", calmar_ratio(e, r, f)) print("Sterling Ratio =", sterling_ration(e, r, f, 5)) print("Burke Ratio =", burke_ratio(e, r, f, 5)) if __name__ == "__main__": test_risk_metrics() test_risk_adjusted_metrics()