DislikedGwan, I think of trading as a game of expectancy. To win with any kind of consistency, we must have a system of entry and exit that provides a demonstrable edge that is capable of overcoming costs. Then, provided positions are sized modestly enough to preserve capital during a sequence of losses, and we maintain perfect discipline in trading our method, we are mathematically bound to profit. However, as with any game, just because the probabilities are in our favor, perhaps even handsomely so, does not offer an iron-clad guarantee.
When I started tinkering with systems, I wrote software to generate a stream of OHLC values, simulating different market conditions: trending, ranging, completely random. For example, if the last tick was upward, I could assign X% probability that the next tick would also be upward: if X>50, that is a “trending” market.
Next, I developed an additional layer, to simulate different entry, exit and MM algorithms. Then I had the software trade these methods over the generated price patterns, and tabulate profit, drawdown, number of wins, losses, average win size, longest losing streak, etc over 10 million simulated trades.
As best as I can remember, here are some of the conclusions that I reached:
1. In a trending market, bottom line is improved by letting profits run and cutting losses short. Hence trailing stops work better than scaling out by taking profits at pre-defined targets. The wider the trailing stop, especially when combined with the greater the trend factor, then the greater the overall profit. Scaling upward into a position exploits trending behavior one step further, and further improves profit.
2. In a non-trending market, exactly the reverse applies, i.e. one should take profits early and hold onto losses, because they are more likely to bounce back. Scaling downward into positions, and even increasing position size while doing so (e.g. martingale), further enhances profit. In fact, the greater the likelihood of price bouncing back, the more aggressive one can afford to be in sizing positions.
3. In a completely random market (i.e. no trend bias or counter-bias), the probability of profit and loss is exactly 50/50. If there are costs, then no system can profit indefinitely under these conditions. No matter what rules of entry, exit, MM are devised, eventual wipe-out is guaranteed. However, the smaller one’s position sizes, the slower the descent toward ruin.
4. Where no bias exists, setting tighter stoplosses will reduce loss size, but will reduce win rate, in almost equal proportion. Same applies to moving stoploss to break-even more quickly. Setting profit targets further away from entry improves win size, but reduces win rate, in almost exactly inverse proportion. Conclusion is that there must be a valid technical reason (i.e. market climate, or bias/inefficiency in price behavior) to support positioning of stoploss or profit targets (e.g. support/resistance), for exit to have any bearing on profitability. Anything less than this amounts to curve-fitting.
5. Same applies to entry. Unless there is a demonstrable underlying (technical or fundamental) reason that highlights inefficiency, we are simply curve-fitting.
6. The fixed fractional method of position sizing has the advantage that positions are sized smaller, in absolute dollar terms, as account balance decreases, and larger, as it increases, allowing a positive expectancy system to generate exponential gains. However, it does create the problem of asymmetric leverage, i.e. when one has lost (for example) 20% of one’s account, it requires a 25% increase on the remaining balance to claw back to break-even.
7. Everything else being equal, position size has no bearing on expectancy. All it does is magnify gains and losses in exactly like proportion. It can only be used to increase profit because conditions favor your system. Base expectancy is the product of entry and exit. Consequently, entry and exit ultimately determine the direction of one’s equity curve, and position size its slope. The true measure of system efficiency = dollars won for every dollar placed at risk, across a given time period.
8. The Kelly formula comes very close to optimizing position sizes, in terms of this efficiency, in the hypothetical situation where account wipe-out is not an issue. In other words, to maximize profit relative to drawdown, without giving wipe-out special consideration. However, given average win rate, win size and maximum allowable drawdown (which is subjective), a Monte Carlo approach is as good as any in calculating optimal position size.
9. Increasing or decreasing position size, depending purely on whether one is currently in a winning or losing streak, has negligible overall effect. However, increasing position size when conditions favor ones entry/exit algorithm (e.g. when using a trend-following method in a trending market), and/or decreasing size when conditions are unfavorable, has a hugely significant effect.
10. The larger the sample size, the more statistically valid the results. (However, this doesn’t take the possibility of obsolescence into account).
11. The corollaries of the above are valid. For example, if letting profits run and cutting losses short tends to lead to greater profit, then we may assume that prices are trending (see point 1) more often than not.
12. Where (hypothetically) costs are irrelevant, trading more frequently (i.e. shorter time frame) generates a smoother equity curve. Everything else being equal (and assuming a positive expectancy entry/exit method), this will leave return unaffected while mitigating risk. However, costs being a reality, the shorter the price moves being traded, the greater the edge necessary to overcome the costs.
13. Extrapolating point 12, trading simultaneous (or overlapping, time-wise) uncorrelated positions leverages time to increase return without affecting risk.
14. All technical indicators, even oscillators, lag price action. All are ultimately derived from OHLC and are therefore multi-collinear. However, the degree of correlation is reduced (i.e. greater independence is introduced) by comparing multiple waves (i.e. timeframes). For example, if H4 is capturing the next higher wave to M30, then phenomena in H4 offers more reliable ‘confirmation’ of M30’s behavior than simply attaching another technical indicator to M30’s chart.
15. All technical analysis makes the assumption that the patterns of the past are in some way, more likely than not, to repeat themselves in the future.
Of course I realize that the above amounts to nothing more than a mathematical model; in reality prices are driven by sentiment, by “commercials” hedging their businesses, by banks and institutions speculating, by the manipulations of market makers, and a host of other factors. But TA makes us the ‘promise’ that the sentiment of all of the participants is ultimately reflected in the current price. Viewed from the perspective of probability theory, price is merely a stream of numbers. It is also the value that we must ultimately buy and sell at, for sentiment can not otherwise be quantified.
Points 14 and 15 apart, much of the above is general in its nature and is relevant to many gaming applications. However, knowing all of this will not in itself make one a more profitable trader (I am living proof of that!). Why? In my experience, it is no easier to forecast imminent market climate (trending, non-trending, etc) than it is to forecast price direction. This is logical, of course, because climate is a product of price movement, and hence can change at any time, and without warning.
DavidIgnored
Seriously thanks for sharing your analysis. In practice I can echo your conclusions although I have not done the through analysis you have. Basically, I liken Forex to playing chess where you cannot predict with 100% degree of certainty your opponent's next move, but if you have thought through all your options you should be able to predict with 100% degree of certainty what your next move is if your opponent does A, or B, or C or D.
If he does A I'll defend and minimized my loss and/or protect my profit. If he does B I'll advance to the next move of progress. If he does C I'll wait and see what his next move is beyond that. If he does D I'll get out of this position and begin a fresh approach.
We cannot predict how each trading day will go. So we refresh our minds at the start of each day with what our choices are before us and which choices will benefit us most and which choices will harm us.