25 June 2020
At the end of 2020, the decrease in GDP, by 9%, should be weaker than expected, thanks in particular to a recovery in some sectors of manufacturing and market services. A recovery seems to be taking shape, even if it will take 12 months before GDP returns to its 2019 level.
French economic activity will continue to recover, but will remain low as a result of factors relating both to supply, since some sectors are still subject to restrictions, and to demand, like domestic and global demand in a highly uncertain context.
Devoteam’s AI modeling is based on a 22% drop in GDP in May (source: INSEE) and integrates multiple data sets such as credit card transactions, household consumption, transportation usage, etc.
“Modelling shows us that the trough of the wave is behind us. This is not the end of a warning signal yet, since the fall has been extremely violent. Of course, these are estimates, but despite the inherent limitations of these AI models, they can be used to make simulations based on various assumptions to model the evolution of the economic activity,” comments Aymen Chakhari, Devoteam’s AI Director.
These relatively optimistic conclusions are based on the absence of a second wave of Covid-19 this winter. The coming weeks, particularly from the start of the new school year, will be crucial for companies: their short-term cash flow capacity and longer-term liquidity will be crucial for their survival.
The AI estimation defined by Devoteam has been trained on INSEE , Statista , Banque de France and OECD data sets. It embeds hybrid predictive models composed of classical decision tree and regression models, combined with conditional symbolic AI under constraints.
This study is based on a thorough and detailed analysis of the following data:
- Information on credit card transactions, statistics from search engines, etc.
- Information on the CAC 40 companies’ production
- Market price data
- GDP evolution until May, 28th: INSEE data
- Household consumption
- Rail freight’s circulation on the SNCF network
- Businesses’ electricity consumption
- Regions’ sectoral structure
Indeed, this study is based on a pure Machine Learning approach coupled with statistical tests carried out on data collection. The data scientists performed iterative feature engineering operations on the data to weight the impact of the chosen parameters more adequately. They ensured the quality of the data used, which is very important for the predictions of an economic context. Numerous time-consuming data preparation operations were carried out to qualify the basic data and thus improve the model’s results.
The distinctive characteristic of this study is that we are in a non-stationary context with a rapid evolution of the activity. Therefore, the reinforcement learning algorithms have been optimized to adapt to this evolution kinetics and take it into account for the predictions. This enables to refine the predictions and improve their accuracy since the models are self-adaptive, unlike purely statistical models which are fixed and highly supervised by history and statistical parameters. This is an undeniable operational and tactical advantage.
We are inviting Individual Blogger | News Publications | Professionals | Brands | Citizens | Kids | Youth | Social Workers to
Contact/WhatsApp: +91 7827887720 Email: email@example.com
Disclaimer: The perspectives communicated in the article/news/press release above are those of the writers’ and don’t really speak to or mirror the perspectives of this website or its personnel. Except if in any case noticed, the writer is writing in his/her own ability. They are not expected and ought not to be thought to speak to legitimate thoughts, perspectives, or strategies of any individual, Institution, organization or foundation.