Build vs Buy: Your Firm's Dream Quantitative Research Platform

Accurately backtesting fees, spread, slippage, in a point-in-time fashion requires a minimum of 6-36 FTE months of engineering - QuantConnect has been constantly improving its backtesting engine since 2012 with a team of ten engineers.

Minute resolution bar data is the most common choice for strategies with a holding period of at least a few hours. Like all intraday datasets, care is required to create bars matching other data sources, and depending on the research application, you may incur expensive live trading data fees and technology overheads. Storage and compute costs of minute data scale dramatically for large markets like US equities with 30,000 unique entities.

Equities data procurement requires an initial purchase, ongoing updates to keep the data up to date, and an expensive live feed if doing intraday trading. Modeling Equities data is difficult, with more than 30,000 listed equities and millions of corporate actions such as IPOs, delistings, splits, dividends, and mergers. Equities data storage ranges from 10GB to several TB, and requires almost constant maintenance to ensure the latest corporate actions are applied. Live trading fees start at $2,000 for low-volume proprietary feeds, to $12,000 per month for the full SIP.

Rough, inaccurate modeling often provides a false sense of confidence in the strategy, which can lead to large losses in live trading. Factors like slippage, fees, and spread are expensive to implement but turn winning trades into losing ones. Most commonly, clients attempt this in a Jupyter Notebook but quickly hit limitations and reach out to us.

Spot pricing (one dimension) and derivative assets (two dimensions) are relatively easy to manage and maintain. Features like universe selection require post-processing on top of the dataset to perform screening and avoid selection bias. Daily updates of this data require a data pipeline to ensure the timely delivery of your strategies. Derivatives like US Equity Options can be challenging due to their size (400TB). Care should be taken to timestamp properly and avoid survivorship bias by modeling current and historically listed assets. This requires supporting a dynamic set of assets rotating in and out of your algorithm universe.

Cloud providers seem attractive at first due to the low capital investment, but quickly escalate in cost when applied to quant-finance. The data storage and CPU requirements of quant trading quickly rack up expensive monthly fees. Time is needed to set up a cloud-hosted operation, and scaling it requires deep knowledge of the vendor API. Clouds appear stable, but for quant purposes like live trading, they often restart servers without warning, potentially causing expensive live trading outages. QuantConnect owns and maintains all our hardware to ensure there are no live trading restarts/outages for the community.