Astignes Capital is an Asia-focused macro and relative value hedge fund with $2.5 billion AUM.

I developed and backtested a multi-factor statistical arbitrage FX trading strategy. I constructed factors using principal components and technical indicators for spot exchange rates, and other inputs like implied rates and macroeconomic risk. I generated trading signals using random forest and logistic regression. The trading strategy, which had a configurable holding period, pairs traded and rebalancing frequency, yielded a backtested out-of-sample Sharpe ratio of 1.3.

I also collaborated with the front office to build a dashboard in React to visualize and track PnL, trading signals, and performance metrics relevant to traders, like volatility and maximum drawdown. Additionally, I documented all the functions I had written.

AI Palette is a predictive analytics SaaS platform for F&B companies.

For my main project on time series forecasting, I developed an Encoder-Decoder Long Short-Term Memory (LSTM) model to predict values for time series trends in the company’s proprietary data on market engagement, and was able to reduce prediction error from the existing model in production by a factor of 3, and reduce training time by a factor of 10. This was made possible by an adding an extensive pre-processing layer, which included STL decomposition, and tuning the hyperparameters of the model to get the best predictions, and by grouping similar time series data and training the model on these groups to reduce training time. The model was selected to be deployed on the company’s customer-facing platform.

For another project on time series analysis, I built a framework based on statistical methods, like the Granger Causality test, to identify trends that have shifted from one food category to another. The trend data was pulled from the company’s servers using Elasticsearch queries, manipulated using pandas, and the statistical tests were conducted using the statsmodels library on Python.

I also redesigned and debugged parts of the existing extract, transform, and load (ETL) pipeline.

SBI Funds Management is a mutual fund with $58 billion AUM.

My project was on alternative data in finance. I conducted extensive literature review on existing ways in which alternative data is used to model macroeconomic indicators.

I scraped over 10,000 unique data points from the web using BeautifulSoup4 and Selenium on Python. This data included real-time and historical pollution levels, Play Store search results, and airline prices.

I recommended 5 stocks on invest in, based on insights gained through regressions performed using collected data as proxies for macroeconomic indicators. For example, the pollution data collected had a positive correlation with the manufacturing activity in the region, and was used as a proxy to get real-time information on the performance on manufacturing companies operating in that region.