The amount of losses is the only parameter which you can control in your trading.
(page xvii)
Nearly all successful speculators who manage to survive in the market prioritize minimizing losses above all else. Traders who solely focus on profits and overlook potential losses can't sustain themselves in the long run. Losses are, essentially, another way of referring to risks.
The main properties of trend-following trading strategies:
1). The percentage of profitable trades is low (36.5%).
2). The overall gains of the system result from the high ratio of average win/average losing trade. The average winning trade is $846, which is bigger than the average losing trade $436 by a factor of two.
3).The average time in winning trades is about three times longer than the average time which the system stays in losing trades (62 bars versus 24 bars)
(page 38)
Trend-following systems don't achieve a high win rate (36.5%). The reason these systems succeed is due to a favorable risk-to-reward ratio (2:1). Another issue is that the time taken for an uptrend is longer than that for a downturn, indicating traders need considerable patience. For many systems, increasing the win rate is challenging, so the focus should be on optimizing the risk-to-reward ratio. This requires cutting losses and letting profits accumulate, aligning with the fundamental principle of trading on the right side of the trend.
The open question remains: at which point does the development and selection phase of a system end and the optimization of your system start?... So the key question for you as a system developer is always: which parameter do you choose from your back-tests? Which settings are likely to continue to produce profits in the future in real trading? The answer to this question is different for each trading system but one rule holds true for all: the neighborhood of your chosen system parameters must be nearly as profitable as your chosen system parameter and the bigger this profitable parameter range is the better.
(page 40)
The essence of this statement is that the condition must hold within the chosen parameter range. Furthermore, the larger the region where this condition approximately holds, the more likely the real-world performance of the system matches its backtesting results. In essence, excessive optimization should be avoided. If machine learning (such as reinforcement learning) is employed for decision-making, relying solely on backtesting is insufficient. Comparing the tables on page 37 and page 49: for the 10/30 dual moving average system, the win rate is 36.5%, the risk-to-reward ratio is 2, and the 6-year return is 7 times the initial investment. However, for the 1/44 dual moving average system, the win rate drops to 26.4%, the risk-to-reward ratio increases to 3.3, and the 6-year return becomes 11 times the initial investment. Therefore, optimizing the win rate is less effective than properly managing stop-loss and exhibiting the patience required to wait for profits. (page 40)
The backtesting of the trading window on pages 50-51 showcases a fascinating phenomenon. It demonstrates that the same strategy yields vastly different results depending on the specific time period within a day. This observation could potentially be applied to the backtesting of several systems currently under development, providing valuable insights for optimizing strategies.
The system's market exposure with the inserted stop was reduced for the first time. Whereas without any exit in place the system was in the market 100% of the time, this risk exposure is now reduced to 73%.
(page 61)
Viewed from a portfolio perspective, reducing market exposure while maintaining the same level of returns effectively results in an indirect increase in future Sharpe ratio
Market behavior is presumed to be a combination of systematic behavior (recurring patterns) and random noise. It is always possible to improve the fit of a rule to a given segment of data by increasing its complexity. In other words, given enough complexity, it is always possible to fashion a rule that buys at every market low point and sells at every market high point. This is a bad idea. Perfect timing on past data can only be the result of a rule that is contaminated with noise. In other worlds, perfect signals or anything approaching them almost certainly means the rule is, to a disturbing degree, a description of past random behavior (i.e. overfitted). Overfitting is manifested when the rule is applied to the test data segment. There its performance will be worse than in the training data. This is because the legitimate patterns found in a training set recur in the test set, but the noise in the training set does not. It can be inferred that profitability in the training set that does not repeat in the testing set was most likely a consequence of overfitting.
page 94. This citation can also be found in David Aronson's book "Evident-based Technical Analysis"
Trading rules should not be overly complex, as complexity can hinder a clear understanding of their essence. The essence of the trading rule is essential for having confidence in the future trades of an optimized system.
Within a single trading system the simplest form of risk management is the control of the distance between the entry and the exit. As a practical example see in Chapter 3.5 how different stops and targets (risk stop, trailing stop, profit target) are added to an entry logic and how this change the risk figures like maximum drawdown, average losing trade and largest losing trade. This step is the first in the risk management and it is the implementation of the quantified risk values into your trading strategy. The second level of risk management is the permanent measurement of risk during the active trading. By screening your trading systems regularly, you have to check if they continue to behave in reality as calculated during testing. You have to watch carefully if markets change concerning point value or volatility.
(page 160)
Risk management is arguably the most vital aspect of trading. Once capital is in play, risks inevitably arise. The first layer of risk pertains to individual trades, with risk primarily existing within the duration from entry to exit. The second layer of risk pertains to overall performance metrics, such as win rate, maximum drawdown, average loss, and more. Correspondingly, risk control methods can be categorized into two types. For the first category of trades, tools like stop-loss, trailing stop, and profit targets can be employed. For the second category of trades, global risk control measures can be introduced (for instance, trading is halted for the day if the drawdown exceeds a certain percentage).
Whether the profit factor is 1.5 or 1.001 is not that important. As long as it is higher than 1 and the system makes steady profits a good money management can improve the results of your trading system. You do not need to have an extraordinarily profitable trading system to gain money. In the long term it's enough to have a stable strategy with a positive expectancy and proper money management.
(page 161)
Position sizing is another highly important topic. How much capital should be allocated for a given opportunity? One possible approach is outlined in Robert Carver's book "Systematic Trading." The central idea is to employ 10 signals to trigger a long position, with each signal assigned a weight. For the sake of illustration, let's assume equal weights for the 10 signals. If 4 out of the 10 signals indicate a long position, then 40% of the originally planned position size is utilized for the long trade. If all 10 signals indicate a long position, then the full 100% of the initially intended position size is used for the trade.
Let's assume that you are trading multiple assets with two different systems, one trend-following system (prone to exploit trending market) and counter-trend system (prone to exploit choppy markets)…
(page 179)
Diversifying among different strategies serves as a robust complement to diversifying across various asset classes. In other words, diversification across different strategies is almost akin to the Holy Grail of trading.
From our experience there is a common rule of thumb for most trading systems. The more signals a system produces, the less profitable it is! Or from the other side, the best trading systems with the highest profits per trade usually don't give entry signals every day in every market phase. Instead, their signals occur rarely, maybe only one time per month, per year, depending on the system logic and the chosen timescale. This means that with the more profitable systems, you will have long periods without any signals. Periods when you just sit like a fisherman for many hours, waiting for the fish to bite into the fishing line.
(page 190)
Quality stocks are worth the wait, even if it takes years. Truly exceptional trades hinge on the timing of entry. Once the market validates your analysis, holding the position firmly is the only right choice.