250 likes | 317 Views
Explore the application of technical indicators, stochastic processes, and autoregressive models on market profitability. Learn how previously used models can be replaced with more realistic distributions to assess price movements and generate better predictions. Discover the correlation between price changes and various technical indicators for informed decision-making and improved trading strategies. Dive deeply into MACD signals' impact on profits and standard deviations, considering new models and Lévy processes for enhanced analysis. Uncover the limitations and inefficiencies of relying solely on technical trends for abnormal returns in dynamic markets.
E N D
Technical Trends: Can they be used to earn abnormal profits? Ryan Weikert
Last Time • MACD • E(profit) ≈ ½ (mu)(S0) • SD ≈ 2/3 (sigma)(S0) • RSI – Quick Trigger • E(profit) ≈ 0 • CCI – Quick Trigger • E(profit) ≈ 0
Technical Indicators applied to random walks generated using Geometric Brownian Motion will not yield abnormal returns
New Studies • More realistic distribution • Lévy Processes • Autoregressive Models (AR & ARMA) • Correlation between price movements and technical indicators
Historical Returns Daily Return
Stable Distribution • rstable(alpha,beta,gamma,mu) • Alpha=parameter • Beta=skewness • Gamma=scale • Mu=shift
Change Model • Old Model • Walk[j]=walk[j-1]*(1+mu*dt+rnorm(nruns,0,s*sqrt(dt)) • New Model • walk[j]=walk[j=1]*(1+rstable(1.9,0,s*sqrt(dt)/2,mu*dt))
MACD with new Model • Standard deviation of closing prices increases ever so slightly while mean remained constant • Expected profit and standard deviation generated by MACD signals remain unchanged • Slight improvement
Lévy Processes: Definition • Starts at some origin at time t=0 • Independent Increments • Stationary Increments • Right continuous with left limits • Geometric Brownian Motion is a Lévy Process
Lévy Processes • Randomize mu • Randomize sigma • Sigma mean reversion
Autoregressive Models • Use previous outputs to predict the next output • Output=Constant+Parameters+randomness(white noise) • AutoRegressive Conditional Heteroskedasticity (ARCH) Models – variance is a function of previous period’s variance
AR Results • E(profit) = 13.85 stderr=.0035 • MACD E(profit)=6.46 stderr=.00234 • Same as before
Correlation between MACD signals and subsequent price changes
S&P 500 Returns after MACD signal For 1 Year Period Beginning N Years ago Days after Signal
Results for S&P 500 • Long Position • E(profit) = 111.78 • Stderr=2.212 • MACD Trading • E(profit)=0.28 • Stderr=1.897
No Abnormal Returns • Efficient Markets • Independent increments • Lévy Processes • Various Distributions • In reality, day to day prices are not correlated • Indicators are lagging • Returns following a MACD signal are not correlated to the signal