Investment portfolio optimizer

This simple script will simulate rebalancing portfolios with given set of assets and display statistics.

How to use

• Install requirements via python3 -m pip install -r requirements.txt

• Save market data into config_returns.csv file. Each row is one rebalancing period, each column is revenue from corresponding asset. Look at example file for details.

• Open config_colors.json and edit asset colors to your taste. Colors are defined by floating-point RGB values in range [0, 1].

• Open config_portfolios.json and add portfolios that you'd like to plot at all times, they will be marked with an X on plots.

• Run optimizer.py with parameters:

  • --precision=10 - Precision is specified in percent. Asset allocation will be stepped according to this value, i.e. each asset will be allocated by multiple of 10%.
  • --hull=1 - Use ConvexHull algorithm to select only edge-case portfolios. This considerably speeds up plotting. In most cases edge portfolios are most interesting anyway. 1 is the fastest, but does not plot too deep into portfolio cloud. If cloud edge is not very well resolved, try higher values. More portfolios will be plotted at the cost of plotting speed. Values higher than 3 are not very useful.
  • --edge=2 - Use number of assets to select edge-case portfolios. 1 will plot only pure portfolios, i.e. havnig only 1 asset. 2 will plot portfolios having up to 2 assets and so on. Values higher than 3 are not very useful.
  • --years=... - specify year selection algorithm:
    • first-to-last - simulate single investment from first to last year in data
    • first-to-all - average of investments from starting year to all later years
    • window-3 - average of all possible 3-year-long investment ranges
    • window-5 - average of all possible 5-year-long investment ranges
    • window-10 - average of all possible 10-year-long investment ranges
    • window-20 - average of all possible 20-year-long investment ranges
    • all-to-last - average of investments from all years to last year
    • all-to-all - average of all possible investment ranges regardless of length
  • --min - Plot theoretical portfolio that allocates only one asset with worst GAGR every year
  • --max - Plot theoretical portfolio that allocates only one asset with best GAGR every year

Check PNG and SVG graphs in result folder for all portfolios performances.

What does it actually do?

Script will generate all possible investment portfolios with every asset allocated from 0 to 100% stepping according to --precision parameter.

Then every portfolio is simulated through market history, rebalancing at every step. Rebalancing assumes selling all assets with new prices and buying them back according to portfolio's asset allocation. For example:

Asset1	Asset2	Asset1 (20%)	Asset2 (80%)	Portfolio value
(start)	(start)	0.2	0.8	1
-10%	25%	0.2 - 10% = 0.18	0.8 + 25% = 1	0.18 + 1 = 1.18
		1.18 * 20% = 0.236	1.18 * 80% = 0.944	1.18 = 0.236 + 0.944
15%	-5%	0.236 + 15% = 0.2714	0.944 - 5% = 0.8968	0.2714 + 0.8968 = 1.1682
		1.1682 * 20% = 0.23364	1.1682 * 80% = 0.93456	1.1682 = 0.23364 + 0.93456

If --hull is specified and is not zero, script will use ConvexHull algorithm to select only edge-case portfolios. Edge cases are calculated separately for each plot. If --edge is specified and is not zero, script will filter portfolios by number of assets, plotting only those that have specified number of them or less.

Notes on averaging

Some graphs display Pop and Dip values for portfolios, meaning largest yearly gain and largest yearly loss. Since year selection algorithm could select more than one year, these values are averaged among all investment periods. For example, consider two 3-year investment periods for asset within 4-year market data:

Year	Asset1	3y-period 1	3y-period 2
2000	-10%	✅
2001	10%	✅	✅
2002	10%	✅	✅
2003	-5%		✅

Two periods have largest dip of -10% and -5% respectively. If there were no dips (all gains are positive), largest dip is considered equal to 0%.

	3y-period 1	3y-period 2
Largest Dip	-10%	-5%

Since there are two periods, average largest dip is (-10% + -5%)/2 = -7.5% That is not the same as 10% which is the largest dip overall, but expected largest dip in a random 3-year period within this year range.

Demo SVGs

You need to download these SVGs to enable interactivity. Generated using --precision=5 --hull=1 --edge=2.