How to win Citi Campus Innovation Challenge Hackathon -By Team Black Pearl from DSE
From participating in the first round for a PPI to finally becoming the National Winners in the grand finale and getting awarded with a huge cash prize in front of the leadership at Citi, it has been quite a thrilling experience.
The competition framework
The 5-day long first round, held in October ’19, was a stock price prediction challenge, while we were expecting a Regression or a Classification problem. Indisputably, we were not prepared for Time Series. We studied a lot on conventional time series algorithms, stock price prediction, deep learning techniques like LSTM and everything which could help us increase our scores even by a few basis points. The level of competition can easily be inferred from the fact that the top 10 teams had made over 2000 submissions with still a few hours left in hand and even for us it took 99 submissions to reach to the top.
After being the winners of the online hackathon, we were eagerly waiting for the offline hackathon. But due to the pandemic, It was impossible to organize an offline hackathon in Pune. Here, I would like to applaud the efforts of the whole team at Citi who organized the finale, and even after so many complications made it possible, and with such ease.
The technicalities
After the gruelling qualifying stage, we knew the finale is going to be extremely tough. It was a live problem at Citi that was given to us, as challenging as it could get, this time we were asked to predict the US Factory Balances with Citi. We were given complete freedom to prepare our training data by including any of the Macroeconomic variables or Financial indicators which may have any predictive power in explaining the variation in Factory Balances.
We tried multiple things like S&P 500 index, Inflation, unemployment, interest, GDP, Industrial Production, Crude Oil Prices etc, but most of these were monthly indexes and we were to predict daily balances. So after our baseline model, we created a monthly index of balances by taking daily balance over monthly average balance. Assuming the business next year will run as usual, we trained our model on monthly average balances and extrapolated the daily balance from the index that we had built before. The leaders sounded extremely interested in this approach, and most of the follow-up questions were also from the indexing method.
Next, we dropped all the monthly indicators, focusing only on the daily data points. We tried to predict the balances using Machine learning techniques. Since all the balances were normally distributed around zero mean, the direction of the predicted balance was crucial. For example, if you predict 5 in place of -5, the MSE will shoot up. So, we made a model to predict the direction of the balance for the next day and used it as a feature in our model. Throughout the hackathon, more than the model, it was the feature engineering that worked for us.
What winning feels like
The submission window was open only for 4 hours till 2:30 PM. You might find it dramatic but there was a technical glitch due to which we were not able to make our submission. It was already quarter past 2 when we made our first submission. We had made so many models, but we couldn’t submit most of them. We had to be selective before making predictions/submission files as time was running faster than usual (of course). Due to the technical glitch, we lost a lot of time, but we were given some additional minutes to make a few more submissions.
We were the first team to present before the panel and so, we barely got a breather before we went into the live chat. The presentation, though unprepared, went better than expected. The depth of questions asked by them along with their interest in our approach was satisfactory enough for us not to care about winning or losing the Citi Campus Innovation Challenge Hackathon.
At 5, as the nail-biting hackathon came to an end and we stepped into the Closing and Award Ceremony, we were just praying for our names to be announced. Finally, it was surreal to see our names flashed on the screen against National Winners. The hard work and the sleepless nights had finally paid off. We are going to cherish these memories forever. Thank you, Citi India, for giving us this opportunity.