0 1
Modeling deviation
Modeling deviation is reflected in two aspects.
First, the model is not perfect. Generally, the quantitative model will be simulated with massive data, but if the historical data used in the test is too small or incomplete, it may lead to the mismatch between the model and the market data.
Second, there are hidden errors in the construction of the model. For example, because the historical data is static and the actual market data is dynamic, if the future function is included in the model, the model will have a good effect when using historical samples for backtesting, but in actual use, there will be signal flicker or price theft, which will lead to great losses in trading.
02
Wrong setting
Missetting risk means that there is an error in setting the relevant parameters of the trading model. For example, in the process of model construction, there may be over-fitting in parameter optimization, resulting in parameter island effect. In addition, if the setting of parameters such as slip point, transaction fee and margin rate is inconsistent with the actual situation, there will be a situation that the quantitative model will gain higher returns in the back test of historical samples, but will lose money in actual use.
03
Execution error
Execution error is a serious risk caused by chaotic error in system program or system architecture. If a trading model should buy when buying signals and sell when selling signals, but before the model finally goes online, the opening order changes from "+1" to "-1", which causes all trading signals to trade in reverse, selling at the price they should have bought and buying at the price they should have sold. The consequences are unimaginable!
In view of this risk, we should not only find out the modeling deviation defects as early as possible through out-of-sample inspection, but also carry out long-term simulation operations on multiple markets or varieties to find out the model defects and avoid applying biased quantitative trading models online.