# Description

Synnax is proud to present the evolution of our financial indicators modeling task with the introduction of Datathon 3! Datathon 3 is a regression problem with multiple targets - a traditional ML problem. All the datasets are conveniently structured - Kaggle style.

In this competition, Synnax challenges participants to predict the financial metrics for the most recent quarter reported for each company. The dataset categorizes each quarter with 141 financial components, such as `TOTAL ASSETS`, `TOTAL LIABILITIES`, `EBITDA`, along with companies metadata and a rich set of macroeconomical indicators. Features names are encrypted for security. Each quarter is prefixed (`Q_0`, `Q_1`, …, `Q_4`), where `_0` denotes the latest quarter in the dataset. The target variables, which were extracted from the `Q_0` statistics, are found in the `targets_train.csv` file.

The task type is identical to the previous competition (Datathon 2), which included the supplementary macroeconomical data as additional potential training features, but we have a few meaningful changes this time:

1. We have revealed the actual dates of financial indicators reported across quarters which are a part of the training features and training targets. There are two new columns in the `X_train` : 
   1. `lastUpdatedQuarterEndDate` which indicates the date when `Q_0` indicators, which are a part of `targets_train.csv` were reported. One can infer the dates when indicators of all the subsequent quarters (`Q_1`, `Q_2`, `Q_3`, `Q_4`) were reported by subtracting three months from `lastUpdatedQuarterEndDate` to get the date of `Q_1` features and so on.
   2.  `lastUpdatedAnnumEndDate` refers to the date of all the features with `Y_0` prefix. Accordingly one may infer that `Y_1` indicators had been reported exactly one year prior.
2. Test dataset: `X_forward_looking` contains the same companies which are a part of `X_train` shifted one quarter forward. It does not introduce any specific data wrangling: just train on the `X_train/targets_train` and predict on the `X_forward_looking` as you normaly would in any other traditional ML problem, but additional explanation may be useful: 
   1. Contestants use as features `X_train` which represent companies' financial indicators reported in `Q_1` - `Q_4` and `targets_train` which represent financial indicators reported in `Q_0`. 
   2. The `X_forward_looking` presents the same structure as `X_train` only the values in `Q_1` - `Q_4` had been shifted one quarter forward. This way the `X_train/Q_1` values are placed in the `X_forward_looking/Q_2`. 
   3. The task is to predict the next quarter's financial indicators in the `X_forward_looking`  dataset.
3. Public and Private leaderboards are introduced for the first time. From the start and throughout the the whole competition contestants will be scored using the private subset of testing data. At the end of the competition the highest scoring (on the Public portion) submissions will be scored agaist the Private part of testing data. 

The **Synnax Datathon 3** task is a highly creative endeavor that can be tackled as a conventional machine learning problem. Participants will train models using a variety of features to predict specific targets. Our dataset offers comprehensive financial performance statistics for each company, derived from components of income statements and balance sheets over the last five quarterly and four annual reports. Additional metadata, such as industry, sector, country, and city, provide further insight into the companies' characteristics.

Contestants may also opt for a time series approach, utilizing values from previous quarters to predict future financial outcomes.

### Competition Timeframe

* Start date: 2nd August 2024
* End date: 12.00pm (noon) UTC 30th August 2024

{% hint style="info" %}
Important note: Synnax has innovated a novel method to assess the probability of default among private companies in the web3 domain, leveraging distributed machine learning efforts from numerous individual contributors. This competition serves as a precursor to the production phase, acquainting participants with the type of data they will encounter and helping to expand our community. Synnax will invite all top-performing participants to join our production pipeline, which includes opportunities for periodic profit sharing.
{% endhint %}