olpsR - On-line Portfolio Selection with R

The olpsR package provides different On-line Portfolio Selection algorithms and functions to deal with the on-line portfolio selection problem where a portfolio is rebalanced in every period to achieve certain goals, e.g. maximizing terminal wealth. Datasets to test portfolio selection algorithms are also included.

For a background on On-line Portfolio Selection see for example [Li and Hoi (2014); http://arxiv.org/pdf/1212.2129.pdf].

Features

On-Line Portfolio Selection algorithms:

• Buy-and-Hold
• Buy-and-Hold best
• Constant Rebalanced Portfolio
• Best Constant Rebalanced Portfolio
• Universal Portfolio (Cover 1991)
• Exponential Gradient (Helmbold et al. 1998)
• Successive Constant Rebalanced Portfolio (Gaivoronski and Stella 2000)
• Anticor (Borodin, El-Yaniv, and Gogan 2004)
• Passive Aggressive Mean Reversion (Li et al. 2012)
• Confidence Weighted Mean Reversion (Li et al. 2013)
• Volatility Timing (Kirby and Ostdiek 2012)

Functions:

• transform asset prices into returns (price relatives)
• transform returns (price relatives) into asset prices
• calculate portfolio wealth of rebalanced portfolios
• transform a sequence of price relatives into a Kelly market sequence
• projection onto a simplex
• ...

Datasets:

NYSE, DJIA, SP500, TSE, DAX

Installation

To install the olpsR package run:

if (!require("devtools")) install.packages("devtools")
devtools::install_github("ngloe/olpsR")

Getting started

Once installed, the package can be loaded using:

library(olpsR)

To test portfolio selection algorithms some return data is loaded using the NYSE dataset. We select two assets, kinar and iroqu:

data(NYSE)
x = cbind(kinar=NYSE$kinar, iroqu=NYSE$iroqu)

Algorithms can be computed by applying alg_ALG on the selected data where ALG is the desired algorithm. For example, to approximate the Universal Portfolio algorithm (UP) type:

UP = alg_UP(x); UP

## SUMMARY:
## --------
##   Algorithm   UP 
##   Assets      kinar iroqu 
## 
##   Terminal Wealth   40.514 
## 
##   Return [%]   28.769              APY [%]   17.944 
##   Risk   [%]   49.758              MDD [%]   82.857 
## --------

Accessing UP then returns a short summary of the algorithm's output. To access the calculated portfolio wealth or the portfolio weights you can type:

UP$Wealth
UP$Weights

The achieved portfolio wealth (performance) can be plotted by:

plot(UP)

To easily compare different algorithms pass them to the plot function:

BH_Market = alg_BH( x, weights=c(0.5, 0.5) )
BH_best   = alg_BHbest(x)
plot(BH_Market, BH_best, UP)

Getting help

For more details and an overview of the implemented algorithms and functions please refer to the package help by typing:

help(package="olpsR")

References

Borodin, Allan, Ran El-Yaniv, and Vincent Gogan. 2004. “Can We Learn to Beat the Best Stock.” Journal of Artificial Intelligence Research 21 (1). USA: AI Access Foundation: 579–94. http://www.jair.org/papers/paper1336.html.

Cover, Thomas M. 1991. “Universal Portfolios.” Mathematical Finance 1 (1). Blackwell Publishing Ltd: 1–29. doi:10.1111/j.1467-9965.1991.tb00002.x.

Gaivoronski, Alexei A., and Fabio Stella. 2000. “Stochastic Nonstationary Optimization for Finding Universal Portfolios.” Annals of Operations Research 100 (1-4). Kluwer Academic Publishers: 165–88. doi:10.1023/A:1019271201970.

Helmbold, David P., Robert E. Schapire, Yoram Singer, and Manfred K. Warmuth. 1998. “On-Line Portfolio Selection Using Multiplicative Updates.” Mathematical Finance 8 (4). Blackwell Publishers Inc: 325–47. doi:10.1111/1467-9965.00058.

Kirby, Chris, and Barbara Ostdiek. 2012. “It’s All in the Timing: Simple Active Portfolio Strategies That Outperform Naïve Diversification.” Journal of Financial and Quantitative Analysis 47 (02): 437–67. doi:10.1017/S0022109012000117.

Li, Bin, and Steven C. H. Hoi. 2014. “Online Portfolio Selection: A Survey.” ACM Comput. Surv. 46 (3). New York, NY, USA: ACM: 35:1–35:36. doi:10.1145/2512962.

Li, Bin, Steven C. H. Hoi, Peilin Zhao, and Vivekanand Gopalkrishnan. 2013. “Confidence Weighted Mean Reversion Strategy for Online Portfolio Selection.” ACM Trans. Knowl. Discov. Data 7 (1). New York, NY, USA: ACM: 4:1–4:38. doi:10.1145/2435209.2435213.

Li, Bin, Peilin Zhao, Steven C. H. Hoi, and Vivekanand Gopalkrishnan. 2012. “PAMR: Passive Aggressive Mean Reversion Strategy for Portfolio Selection.” Machine Learning 87 (2). Springer US: 221–58. doi:10.1007/s10994-012-5281-z.