Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
braverock
GitHub Repository: braverock/portfolioanalytics
 Path: blob/master/sandbox/opt_parallel.R
1433 views
1
library(PortfolioAnalytics)

2


3
data(edhec)

4
R <- edhec[, 1:5]

5
funds <- colnames(R)

6


7
# Construct initial portfolio

8
init.portf <- portfolio.spec(assets=funds)

9
init.portf <- add.constraint(portfolio=init.portf, type="weight_sum", 

10
                             min_sum=0.99, max_sum=1.01)

11
init.portf <- add.constraint(portfolio=init.portf, type="long_only")

12
init.portf <- add.objective(portfolio=init.portf, type="risk", name="StdDev")

13


14
# Run optimization with DEoptim

15
minStdDev.DE <- optimize.portfolio(R=R, portfolio=init.portf, 

16
                                   optimize_method="DEoptim", 

17
                                   search_size=2000, 

18
                                   traceDE=0,

19
                                   trace=TRUE)

20
xtract.DE <- extractStats(minStdDev.DE)

21


22
# Here we extract the objective value returned from the optimizer of each

23
# iteration. I'm not sure how useful this information is.

24
obj.DE <- xtract.DE[,"out"]

25
hist(obj.DE)

26
rug(obj.DE)

27
plot(density(obj.DE))

28
qqnorm(obj.DE)

29
boxplot(obj.DE)

30


31
# Run optimization with random portfolios

32
minStdDev.RP <- optimize.portfolio(R=R, portfolio=init.portf, 

33
                                   optimize_method="random", 

34
                                   search_size=2000, 

35
                                   trace=TRUE)

36
xtract.RP <- extractStats(minStdDev.RP)

37
obj.RP <- xtract.RP[,"out"]

38
hist(obj.RP)

39
rug(obj.RP)

40
plot(density(obj.RP))

41
qqnorm(obj.RP)

42
boxplot(obj.RP)

43


44
# I think the best way is to do a sort of bootstrap by running several

45
# hundred or thousand (depending on your resources) optimizations and

46
# analyze the objective value from each optimal portfolio

47
opt <- optimize.portfolio.parallel(R=R, 

48
                                   nodes=50,

49
                                   portfolio=init.portf, 

50
                                   optimize_method="random", 

51
                                   search_size=2000, 

52
                                   trace=TRUE)

53
opt

54
xx <- summary(opt)

55
obj_val <- xx$obj_val

56


57
# estimate of the objective measures, objective value, and weights from the

58
# optimal portfolio of each optimization

59
apply(xx$stats, 2, mean)

60


61
# plot the objective values from each optimization

62
hist(obj_val)

63
rug(obj_val)

64
plot(density(obj_val))

65
qqnorm(obj_val)

66
qqline(obj_val)

67
boxplot(obj_val)

68


69
# These should match the print method

70
# estimated objective value

71
mean(obj_val)

72
# percentile confidence interval estimate

73
quantile(obj_val, probs = c(0.025, 0.975))

74


75


76