This is a report of the performance of TFadapt, a trend-following system that adapts to changing market conditions, on historical data of 8 different international indices since the start of 1990. The outstanding and statistically significant performance of this system in SPY since inception was detailed in this post where the system was introduced.
Trend-following is in principle very hard because, amongst other things and issues, the duration and magnitude of future trends are random variables and market conditions can emerge where there are not enough profits from trends to cover losses from choppy market action for extended periods of time. The mathematical proof of this fact was given in this post.
In addition, most popular trend-following systems used by traders and funds managers have emerged after backtesting programs were made available in the 1990s and are mostly the outcome of the dangerous practice of data-snooping. In another post I have shown using simulations that the well-known and widely used 50-200 golden cross system lacks intelligence and although it is very popular it was probably an artifact of data-mining of the 1990s.
In principle, trend-following systems attempt to do better than buy and hold. If a system cannot do substantially better than buy and hold in terms of risk adjusted return, which is measured as the ratio of CAGR to max drawdown, then the system is worthless and using it, instead of buying and holding, is an irrational act. Thus, when designing trend-following systems, the following two objectives must be met amongst other things:
(1) The system CAGR must be higher than that of buying and holding.
(2) The system maximum drawdown must be smaller than the one experienced during buy and hold.
The combined conditions above if true result in a higher risk adjusted return, as measured by the ratio CAGR/maxDD, when compared to risk adjusted return of buy and hold.
Performance results for the TFadapt system
Below are performance tables for the CAGR, MaxDD and CAGR/MaxDD ratio for 8 major international indices with data starting on 01/02/1990 or later depending on publication date. Backtested performance is based on an initial capital of $100,000, fully invested in a hypothetical number of index shares calculated as the ratio of available equity to index price. This system is long/short symmetric as short signals exit long positions and the other way around.
Table 1. Return score: 12/16 (75%)
|Index||B&H Return||50-200 GC||TFadapt Return||Score|
From Table 1 it may be seen that the return of the TFadapt system was lower only in the case of STI and RUT, as compared to buy and hold, and in the case of HSI and S&P/ASX 200, as compare to the 50-200 GC . When compared to both the buy and hold and the 50-200 golden cross (GC) systems, the TFadapt system has a score of 75%. (A score of ++ means that the TFadapt system return exceeded both buy and hold and the 50-200 GC system).
Table 2. Max Drawdown score: 16/16 (100%)
|Index||B&H MaxDD||50-200 GC MaxDD||TFadapt MaxDD||Score|
From Table 2 it may be seen that the buy and hold max drawdown was devastating in each and every case, exceeding the 50% level. The 50-500 GC system is not able to get a significant reduction across the board but only for 3 indices but the TFadapt system manages to get significant reductions below the important 50% level in all cases and receives a perfect 100% score.
Table 3. CAGR/MaxDD score: 14/16 (87.5%)
|Index||B&H CAGR/MaxDD||20-500 GC CAGR/MaxDD||TFadapt CAGR/MaxDD||Score|
From Table 3 it may be seen that the TFadapt system has one tie and one negative performance in terms of the ratio CAGR/MaxDD. The overall score is 87.5% in this case, indicating the superiority of this adaptive system as compared to buy and hold and the 50-200 golden cross system.
The TFadapt system is available for sale. A performance report can be produced before a sale is made for up to 10 markets with data supplied by the customer. The results will be reproducible by the customer (no randomness is involved in the system decisions).
Disclosure: no relevant position at the time of this post and no plans to initiate any positions within the next 72 hours..
Charts created with AmiBroker