Suppose you trained a nifty classifier and also showed good results over a blind or live set, now should you start trading? While it is possible, it isn't so easy. The following are some reasons why:
- Historical analyses versus streaming data: This historical data is often cleansed and near perfect, but streaming data does not offer such benefits. You will need code to evaluate the stream of data and throw out potentially unreliable data.
- Bid-ask spread: This is the biggest surprise novice's face. There are actually two prices in the marketplace: the price at which you can buy and the one at which you can sell. You don't both buy and sell at the typical market price you see (that is just the last meeting point on the two, called the last price). Buying a holding and immediately selling it loses money because of this gap...
