Dev Diary 3

Welcome to Dev Diary 3! Today’s topic will be the second market condition included in the demo: Moderate difficulty with an Esoteric theme (Moderate/Esoteric). In contrast to Dev Diary 2, we will walk through the market condition to infer an optimal trading strategy.

This is intimidating. Where to start first?

Since we need to start somewhere, let’s look for similar relationships as seen in last week’s diary on the Easy/Aggressive market condition: do changes in one stock predict the changes of another stock? As illustrated below, these relationships are completely flat (in other words, changes in one stock do not predict changes in other stocks).

However, the screenshot has an interesting characteristic: the changes for all of the stocks are mostly bounded above and below. This indicates that something may be bounding the price of each stock. However, the prices of the stocks are also bounded with no clear relationship.

Given this information, it may be helpful to look at a quick game in Moderate/Esoteric to better understand why prices and changes are being restricted in some way. In fact, prices tend to “jump down” when they get too high (~$130) and “jump up” when they get too low (~$70).

This observation suggests that the price of a given stock may predict the future change of the same stock (instead other stocks). This phenomenon is observable in that stock prices, certainly above ~$130, but also above ~$120 have large decreases on the following turn. Additionally, there is a notable number of simulations with larger increases at lower prices, but it is not a clean relationship because the mean change (that is, the orange line) stays relatively flat even at lower prices. Thus, the “jump up” phenomenon is not entirely explained by price alone.

Can we create a strategy from our current observations?

These observations suggest two principals for a trading strategy:

1. Avoid expensive stocks: It is clear that stock prices have a ceiling effect, which results in a large price decrease, that starts around ~$120.

2. Cheap stocks: Stocks with prices below ~$80 can have notable increases, but it is not entirely explained.

A simple trading strategy for this approach would sell all stocks at the beginning of the turn and buy the cheapest stock, as implemented below.

In the simulation, this simple strategy works well but has a slightly worse average portfolio value than the Market-Specific AI, suggesting room for optimization.

While the nature of the “jump up” mechanic is not clear, the behavior of the Market-Specific AI suggests that buying the cheapest stock occurs only when at least one stock achieves the price ceiling. Since this situation is less common, it would help explain why the mean does not consistently change at lower prices even though some individual points clearly do. Thus, we could buy the cheapest stock only when 1 or more stocks have a price above $120 instead of always buying the cheapest stock. In the other situations, we diversify into three stocks with high dividends (QE, REAL, and NRI) in an attempt to generate marginally higher returns.

The simulation results in slightly better performance than our original strategy and the Market-Specific AI.