Time Series
09/03/2023

Quantmetry's Winning Methodology: Achieving an 8% Profitable Investment in M6 Forecasting Competition


Auteur : Rima Hajou
Temps de lecture : 9 minutes
Quantmetry.com : Quantmetry's Winning Methodology: Achieving an 8% Profitable Investment in M6 Forecasting Competition

We are excited to share our success story in the M6 forecasting competition’s [1] investment category. Our team of Time Series data scientists at Quantmetry developed a winning methodology that not only secured the third prize but also produced an impressive profit of 8% if we had invested according to our strategy.

Final leaderboard of M6 competition, sorted by Rank (investment) – Link

In this post, we will delve into our winning methodology for both the forecasting and investment components of the competition. Our approach was strategic and rigorous, resulting in a solid investment decision-making process that proved successful in the competition. Join us as we share our takeaways and insights from this exhilarating competition.

Our team of Time Series data scientists at Quantmetry are thrilled to share our success in the M6 forecasting competition. After a year of hard work and dedication, we have secured the third prize in the investment component of the competition. The M6 competition challenges participants to forecast ranks for 100 financial assets and this year’s edition was particularly challenging with a live submission schedule every 4 weeks.

So what is the M6 competition?

The M forecasting competition’s 6th edition challenged more than 160 participating teams to forecast the ranks of 100 financial assets, including stocks and ETFs, in real-time. The competition required participants to submit their predicted values every month according to a specified schedule.

The competition was divided into two components: forecasting and investing. Participants had to submit exactly 12 forecasts, with each submission being required every 4 weeks. The forecasting component required the forecast of the rank probability of each financial asset, with returns in the 1st, 2nd, 3rd, 4th, or 5th quantile. The investment component involved making decisions to invest or not invest based on the forecasting probabilities estimated in the first component.

In this post, we will provide a summary of:

  1. Explaining the forecasting component
  2. Our approach for tackling the forecasting component
  3. Explaining investment component
  4. Our approach for tackling the investment component
  5. Takeaways

Let’s start!
(if you already know the guidelines [2] of the competition you can skip part 1 and 3 😉 )

1. Explaining the forecasting component

In this section, we will explain in detail the objective of this component.

The objective is to submit the rank probability of each of 100 assets monthly.
These 100 assets represent:

  • 50 stocks from the Standard and Poor’s (S&P) 500 index
  • 50 international exchange-traded funds (ETFs)
A glance of some of the stocks

Figure 1 – A glance o some of the stocks

 

The 50 stocks and 50 ETFs are selected such that they are broadly representative of the global market.
The stocks represent several sectors (Financial, industrial, health care, technology, real estate, consumers good and energy) and the ETFs represents Equities, Fixed Income, commodities in different markers.
We will start by explaining step by step how to estimate the rank probability of each asset.

Step 1: For each asset, predict the stock return in 4 weeks.

Figure 2 shows the value to predict, as well the values that we can use to estimate this prediction. Noting that we can use all the historical data we want, and not exclusively those of the asset we want to predict.

The value to predict is the asset percentage change (stock return):

    \[ stock\;return = \frac{Price_{J+28} - Price_J}{Price_{J}} \times 100 \]

Figure 1- Timeline of features vs value to predict

Figure 2- Timeline of features vs value to predict

Step 2: Now we have the predictions of stock return of each of the 100 assets, we will sort them from smallest to largest according to the values.

The 20 assets with the smallest values (worst performing) will be in rank 1.

  • The second 20 assets with the smallest values will be in rank 2.
  • The third 20 assets with the smallest values will be in rank 3.
  • The fourth 20 assets with the smallest values will be in rank 4.
  • The fifth 20 assets with the smallest values will be in rank 5 => the 20 assets with the largest values.

The figure 3 shows an example of how the ordering of ranks is done.