Conflicting information | Traders Hideout


futures.io - futures trading strategies, market news, trading charts and platforms


Traders Hideout


Discuss day trading practices and futures trading strategies on this forum for all markets. This forum is also for discussing and reviews for brokers, data feeds, and commercial or third party add-ons




 

Conflicting information

  #16 (permalink)

North Carolina
 
Trading Experience: Beginner
Platform: NinjaTrader, Tradestation
Favorite Futures: es
 
Posts: 644 since Nov 2011

Let me play devil's advocate and suggest some reasons that traditional system trading may not be the solution:

Pattern recognition
Humans are still generally far better at pattern recognition then computerized methods especially methods that are easily/readily available in most trading platforms. Deep learning is certainly improving this area.

Cognition
Humans can think. Algorithms merely capture a specific line of thinking but do not think themselves. Computers are basically Turing machines but human cognition is nothing like that. Humans have achieved far more then AI has ever achieved.

Repeat-ability
Trading systems are primarily based on the idea of repeating patterns. However, we know that some of the best times to profits in markets are the result of unique factors and events. We, also, know that the HFT spend enormous amounts of money and energy to buy access to the fastest time frames suggesting that truly consistent and robust repeatable event trading may not exist beyond the smallest time frames. If just as profitable and repeatable patterns could be found on higher time frames then why would they invest so much in getting the fastest connections, speeds, etc?

But, generally I do agree trading systems do offer many other types of advantages. The solution, as such, is probably to focus on market cognition and ways of quantifying the risk or what is quantifiable while accepting the uncertainty inherent in trading.

Accepting market cognition can not be captured in any singular system
It is probably impossible to capture human-like market cognition in any singular system and even if it did the method would likely be as blackbox as discretionary trading. However, some assortment of systems could start to simulate such a thing. This suggests that one should focus not on building the best or greatest system but rather building a set of systems established on understandable principles.

Moving toward mixed graybox style solutions
Using the backtest to help quantify the risk and position size and the exact entry while leaving aspects of the prediction and pattern recognition to the trader. This type of logic can be summarized as: It is prerequisite but insufficient to justify taking a trade merely because it was profitable in the past. As well, it is is prerequisite but insufficient to take a trade merely because one has an speculation about what the market will do in the future. Rather, you need both a historical validated conception in the past and an intuition about the future.

Building different types of systems that do not rely (as much) on past market data
This would be moving toward cognitive systems. It is a bit difficult to imagine exactly how such a system might look or behave. However, some examples might look like, if a model suggest mean reversion is going to be higher in the market with a given level of volatility then even though a particular market may not have expressed any mean reversion in the past (historical prices) that one would trade it using a mean reversion paradigm should certain predictive or indicative factors come into play, such as higher volatility. This is a simple example, obviously but illustrates the concept of how trading systems could be built in a quantitative way without relying on historical data. Such systems can trade the far right edge of the possibility, they are able to trade patterns that have never been expressed in past data.

Another example of a cognitive system might be to build a strong quantitative model of where oil prices are likely to be over the next 6 months and if they fall outside that level to trade with the breakout on the expectation that other traders bet on the quantitative model. For example, if the best quantitative models predict oil prices to stay below $60 and oil prices rise above $60, you buy oil because you know that other traders are likely caught off guard which should result in at least some temporary continuation of the move. The problem is if your models are good then you may have to wait a long time for the models to be wrong. In this way, you are inferring the actions of the other participants.

Reply With Quote