Quant And Machine Learning Portfolio Construction: Mimicking An Asset Price Series With Nearest Neighbors

We will use a specific example to illustrate how to create a portfolio that mimics the time series of prices of a given asset. Here, we will mimic abrdn Physical Platinum Shares ETF (PPLT) with a stock portfolio for which the weights are proportional to 1/distance.

For 126 days of data, we use both fractional return and zscore paa (piecewise aggregate approximation) data representations and dynamic time warping distances (See Dynamic Time Warping Nearest Neighbors). For N stocks, we compute distances, sum their reciprocals, then weights are equal to reciprocal distances divided by this sum. The procedure yields the following portfolios with 15 stocks with 126 day average volume >= 100,000.

Fractional Return PAA

Zscore PAA

Here we chose an asset for the current time period. However, this is clearly not necessary. We can choose any asset for any past time period which implies a classical charting view of portfolio construction. We must choose the parameters mentioned above to match characteristics of the time series of prices that we deem important.

If you find such information useful, we ask that you provide value in return via the Value 4 Value links on the homepage. We would also appreciate you spreading the word about our work.