A research backtest is a fast, approximate test of a trading idea against historical data, carried out at the exploratory stage of strategy development. It is usually vectorized: signals and returns for the whole history are computed in one pass over arrays of data, with simple assumptions such as trading at the closing price and a flat allowance for each transaction cost. The purpose is to screen many ideas cheaply and discard those with no promise.
The term is used in contrast with an event-driven backtest, which replays market data one event at a time and models orders, fills and delays in detail, at the cost of more code and slower runs. Because a research backtest simplifies execution, its results tend to be optimistic, and its speed makes it easy to try so many variations that overfitting results. It is also prone to look-ahead bias if data are not aligned carefully in time.