In the past 20 years, equity markets have increasingly shown a boom and bust behavior. In such a market settings the ability to forecast bubbles formation and subsequent crashes or the beginning of a market rebound would be of great value. Attempting to forecast the formation, end of bubbles and subsequent changes of regime, Prof. Sornette and co-workers developed a rational expectation bubbles model called Johansen Ledoit-Sornette model (JLS) [6] [7] [8] and based on complex systems physics.
The core of the JLS model states that, in a bubble formation, prices are following an unsustainable faster-than-exponential growth punctuated by log-periodic accelerating oscillations. This price behavior comes from the fact that traders are forming a complex network of relationships and are influencing each other in their decision making. In a normal market the number of traders willing to buy and the number willing to sell usually balance each others’ out and leads to an unordered and fluid market. Individual opinion wins over group opinion on which action to take (buy, sell, hold). However in a bubble formation, big groups of non-directly connected traders start to influence each other, the word spreads in the network and creates big waves of traders willing to act the same way. This leads to strong positive feedbacks. Eventually a crash (resp. rebound) occurs when order has taken on the network and most traders want to sell (resp. buy) at the same time the same asset.
This market behavior is intrinsically transient. Because of this transient aspect, some predictability arises which is attempted to be captured in the JLS model. The JLS model predicts the market price’s behavior only up to the bubble termination. After this termination the price regime can change to anything from a gentle deflation of market prices to a violent crash. This behavior has strong similarities with complex systems in physics like earthquakes, avalanches or crack propagation. Same tools and approaches are therefore being used to understand market behavior and try to predict any crack in the market.
CONTENTS
1 introduction
2 the jls model and bubble conditions
2.1 Idea
2.2 The model
2.3 Implementation
3 pattern recognition method
3.1 Idea
3.2 Sub-intervals construction
3.3 Extremums definition
3.4 Procedure in details
- 3.4.1 Class
3.4.2 Group
3.4.3 Informative Parameters
3.4.4 Questionnaire
3.4.5 Traits
3.4.6 Features
3.4.7 Alarm Index
3.4.8 Prediction
3.5 Alarm Index Results
3.6 Error diagram
4 trading strategy
4.1 Idea
4.2 Influence of qualify choice
4.3 Results
5 the informative parameters bottle-neck
5.1 Learning set length
5.2 Class I width
5.3 Sub-interval step
6 conclusion
a appendix
a.1 S&P 500
a.2 Nasdaq
a.3 Russell 2000
a.4 FTSE 100
a.5 CAC40
a.6 SMI
a.7 DAX
a.8 Nikkei 225
a.9 Hang Seng
a.10 ASX
bibliography
