Table 51 does this calculation for you, if your instruments come from multiple asset classes. The relevant values for instruments drawn from a single asset class are in table 52.
Table 51: Maximum account level risk target when both instruments and trading rules are added to the Simple System, where instruments come from multiple asset classes Multiply the proportionate increase in Sharpe ratio for instruments (from table 44 in chapter seven) by the relevant ratio for rules (table 50). Values are then rounded to nearest %, and lim ited to 30%.
Table 52: Maximum account level risk target when both instruments and trading rules are added to the Simple System, where instruments come from the same asset class Multiply the proportionate increase in Sharpe ratio for instruments (from table 45 in chapter seven) by the relevant ratio for rules (table 50). Values are then rounded t o nearest
%.
Remember, from part two, page 99 , that your risk target should be the minimum of calculated values related to: (i) allowable leverage, (ii) prudent leverage, (iii) your appetite for risk,
and (iv) the expected Sharpe ratio of your system. Adding more rules or instruments only affects the expected Sharpe ratio. You should check that the other calculated values don’t give you a lower risk target.
Also bear in mind that the figures in table 51 are account level risk targets ; with more than one instrument in your account you need to multiply these by the instrument diversification multiplier (IDM) to get the risk target for individual instruments, as I explained in the previ ous chapter.
For example, suppose you are trading three instruments which are drawn from multiple asset classes, with five trading rules. I already calculated the account level risk target: 18%. The IDM
for three instruments across multiple asset classes is 1.22 (from table 43 in chapter seven). So, the correct instrument level risk target is 1.22 × 18% = 22%.
What criteria do we use for selecting and weighting trading rules?
Which rules and which weights?
How should you choose your opening rules, and which weights should you use when averaging them to work out your opening position? Potentially important criteria are: (a) pre-cost performance, (b) capital required, (c) time constraints, (d) different cost levels, and (e) how much diversification t hey provide.
In fact, we can safely ignore both (a) pre-cost performance and (b) capital requirements. Figure 20 near the start of this chapter suggests that we can’t distinguish the pre-cost performance of different trading rules. The number of instruments you can trade simultaneously is limited mainly by your capital ; each additional instrument needs its own capital allocation. But no such constraint exists for multiple trading rules, as they share the same pool of capital. The same minimum capital is required regardless of whether you are using one opening rule o r a hundred.
A more serious problem is the extra time needed to run additional calculations. Some of the trading rules in this chapter don’t need much additional effort: moving averages, and breakouts. If you use spreadsheets to calculate the value of your moving averages, then you can set up multiple sheets, each linking to your price data. You can then add extra sheets for different breakout rule variations. If you are using an internet data provider to calculate your moving averages or breakout signals, then you can bookmark multipl e web pages.
Carry is more work as you need to gather extra data. For spot FX
you’ll need interest rates; whilst for CFD, spread bets and futures you require a spot price, or the price of another traded
contract. Margin traders will need interest rates, borrowing fees, and divi dend yields.
How do costs and diversification affect the choice of rules and weights to use when averaging? Let’s assume for the moment that you have sufficient time to handle numerous trading rules. You decide to include six moving average crossovers (from 2,8 to 64,256), six breakouts (from N=10 to N=320), and the carry rule.
How would you set y our weights?
Firstly, you should exclude any trading rules that are too expensive . Discard any rule with risk-adjusted costs that are greater than one-third of their expected pre-cost Sharpe ratio.
20
¹¹78 For the six instruments I used in part two to illustrate the Starter System, here is a list of trading rules that are too expensi ve to trade:
Here are the rules that I would keep:
Corn future: Moving averages 2&8, 4&16, 8&32, 16&64, 32&128, 64&256. Breakouts N=10,20,40,80,160 ,320. Carry.
Euro Stoxx CFD on future: Moving averages 8&32, 16&64, 32&128, 64&256. Breakouts N=40,80,160 ,320. Carry.
Gold spread bet on future: Moving averages 8&32, 16&64, 32&128, 64&256. Breakouts N=20,40,80,160 ,320. Carry.
S&P 500 margin account: Moving averages 8&32, 16&64, 32&128, 64&256. Breakouts N=20,40,80,160 ,320. Carry.
AUDUSD: Moving averages 8&32, 16&64, 32&128, 64&256. Breakouts N=20,40,80,160 ,320. Carry.
If you are trading multiple instruments you could end up with a different set of trading rules for each instrument. This might sound too complicated. If you prefer using the same rules for all your instruments, then implement those that are viable for the most expensive instrument you are trading. For example, if you are trading corn futures and Euro Stoxx CFD, then you would use the set of rules which worked for Eur o Stoxx CFD.
Now we have chosen which rules to use, what weight should we use?
You should give more weight to the most diversifying trading rules . I recommend doing this with a top down allocation method.
20
¹¹79 First, we allocate our weights equally across different styles of trading. Trend following is one style (which includes both moving average crossovers, and breakout); whilst carry is another style. I’d allocate half my weighting to trend following, and then another ha lf to carry.
Since there is only one carry trading rule that will get the entire 50% weight. The other 50% I’d split equally between the two types of trend following rule: moving average crossovers
(25%), and breakout (25%). Finally, I’d split rule allocations equally across variations of each rule (different length 20
crossovers, and different values of N for breakout). ¹²70
For example, consider the weights for gold spread bets. We put half into carry (50%) and split the other 50%: 25% into moving average crossovers (MAC), and 25% into breakout. The 25% for MAC
is divided four ways, as there are four MAC rules which are cheap enough to trade. Each gets 25% ÷ 4 = 6.25%. In the breakout allocation there are five rules we can allocate to, since breakout 10 is too expensive, so each gets a weight of 2 5% ÷ 5 =
5%.
Table 53 shows the calculation of weights for the six example instruments.
Table 53: Weightings to use when averaging trading rules for different instruments
For each instrument allocate half to carry, and half to trend following. Then split the trend following allocation: half to moving average crosssover (MAC), half to breakouts. Then divide the MAC and breakouts between the variations of these rules that are cheap enou gh to trade.
If you are short of time then you can drop the Carry rule, which is very time consuming to calculate. You should double all the remaining values shown in this table for moving averages and breakout rules to get the appropriat e weighting.
Should you change your closing rule?
If you use different opening rules, do you also need to modify your closing rule? The closing rule is a stop loss level calculated as a fraction of instrument risk, and this fraction is calibrated to match the typical holding period of the opening rule. In the Starter System I set this fraction at 0.5, which corresponds to the expected trading frequency of the 16,64 moving average crossover rule. If you start using opening rules with a shorter or longer trade horizon, then you need to recalibrate your c losing rule.
Firstly, calculate your expected number of trades , as a weighted average of the trading frequency of your opening rules and the weights y
ou have on those rules. For example, suppose you are running a system with 25% in the 16,64 moving average crossover (5.4 trades per year, from table 8 ), 25% in breakout with N=80
(14.8 trades per year, from table 48 ), and 50% in Carry (5.4
trades per year, from table 49 ). The weighted average number of trades per year is:
(0.25 × 5.4) + (0.25 × 14.8) + (0.5 × 5.4) = 7.75
This is a little quicker than the Starter System, which with just the 16,64 moving average rule is expected to place 5.4 trade s each year.
Now consult table 54 to find the most appropriate volatility fraction given your tr ading speed.
Table 54: Faster trading needs tighter stop losses All values calculated from back-testing different MAV opening rules and stop loss fractions over the instruments in my data set. Row in bold is Starter System. Figures are copied fr om table 12 .
For an opening rule with 7.5 trades annually the closest matching fraction would be 0.4. If you are using the recommended rules and weights from the last section (set out in table 53 ), then you can use table 55 to find the appropriate volatility fraction for your c losing rule.
Table 55: Trading speed and recommended stop loss volatility fraction if using suggested trading rules and weights from table 53
Number of trades per year calculated as a weighted average of trades for each rule (from tables 8, 48 and table 49) using weights from table 53. Suggested volatility fraction from table 54, with interpol ated values.
Does it ever make sense to add or remove trading rules?
It is absolutely fine to add new trading rules to your system at any time. Just use the new set of opening rules when you are ready to open a new position, and don’t forget to update your weighting scheme. What about removing a tr ading rule?
In theory, if you find that a trading rule is losing money, and those losses are statistically significant, then you should remove it from your trading system. In practice it is extremely unlikely that your losses will ever be large enough, or that you will trade for long enough, to reach that point. Tracking the performance of every individual opening rule is also extremely difficult and time consuming, unless your system i s automated.
I recommend: do not remove trading rules from your system .
Using multiple trading rules in practice Before you start trading
There is a bit more work to do before you begin trading with multiple tr ading rules.
Here is the updated trade plan for mul tiple rules:
Running the system with multiple opening rules The only change when operating the Starter System is to the opening rule: when considering whether to open a trade you need to check each trading rule. Once you have the buy or sell decisions from each rule then you need to come up with a single buy or sell decision, using the weighted average as I discussed back on page 210 .
¹¹¹ This test was done using the trading rules in my own system, which comprise the 13 rules discussed in this book, plus some additional rules which I discuss on my blog: qoppac.blogspot.com
.
¹¹² For reasons of space I can’t show the improvement if you started with the Starter System and two (or three, or four...) instruments and then added further rules (without adding additional instruments), as these will be a little different depending on exactly which instruments you started with. But the overall pattern in expected improvement will be the same, regardless of how many instruments you’re trading when you start to add rules.
¹¹³ At least for pre-cost returns. As I’ll discuss later in the chapter some rules trade too quickly to make them viable for certain instruments.
20
¹¹74 My back-testing results confirm that N=10 (about two weeks) is about as fast as we can successfully trade the breakout. I also found that repeatedly doubling the value of N gives a set of breakouts that aren’t too similar. Beyond N=320 the rule starts to trade too slowly; its profitability falls and it becomes too much like a constant long position in many instruments.
20
¹¹75 We could use the spot price here, but this example is included for completeness as there are many futures with no spot price, such as government bonds.
20
¹¹76 The “Turtle Traders” were a group of trading novices who were taught how to use a simple trading system and given capital to trade it. Many then went on to become professional traders. See The Complete Turtle Trader by Michael Covel.
20
¹¹77 I use ten here to be consistent with a concept I introduce in part four. In practice you can use any number you like here.
20
¹¹78 See page 78 .
20
¹¹79 There is much more on this technique in my first book, Systematic Trading .
20
¹²70 Because we drop variations that are too expensive (and exceed the ‘speed limit’ of one-third of expected pre-cost Sharpe ratios spent in costs) this will always lead to allocations that trade
slower than the ‘speed limit’. This is conservative, but sensible given the uncertainty in pre-cost Sharpe ratios.
Part Four: Advanced Trading
Chapter Nine
From Discrete to Continuous Trading
“You’ve got to know when to hold ‘em. Know when to fold ‘em. Know when to walk away. And know when to run.”
Lyrics from the Kenny Rogers song ‘ The Gambler ’ written by Don Schiltz in 1976
Kenny Rogers was singing about playing poker, but these lyrics apply equally well to trading. Knowing when to fold – when to close an unsuccessful trade – is a vital part of any trading strategy. The Starter System introduced in part two, and developed in part three, does this with a stop loss.
The Starter System has a pretty simple structure. We use one or more trading rules to decide whether we should be long or short the market. Then after a position has been opened, we wait for our stop loss to b e triggered.
Pretty simple, but we can make it even simpler: by dropping the stop loss . This may sound like dangerous lunacy, and no doubt Kenny would be very sceptical. But the Starter System doesn’t really need a stop loss.
Why it can be safe to trade without a stop loss Why do we have stop losses?
They control your risk, by limiting how much money you can lose in any indiv idual trade.
They get you out of losing trades before they ge t any worse.
They seem like pretty important jobs, but they can be covered by other parts of the Starter System. Stop losses are effectivel y redundant.
Firstly: risk control . Stop losses provide trade level risk control . The stop loss used in the basic Starter System works out to a maximum capital loss of 6%, on any given trade. We can work this out using formula 25 :
Capital at risk per trade % = Annualised risk target % × stop loss fraction
= 12% × 0.5 = 6%
But the maximum loss depends on how long we are expecting to hold positions for, since this in turn determines the stop loss fraction. The capital at risk on each trade isn’t consciously set in the Starter System. It is just a value implied by the stop loss fraction and our risk target.
In the Starter System the annualised risk target is the primary method of risk control. We design our system so we know how much we expect to make or lose on any given day, and that determines how large our positions should be. This is time-based rather than trade-based risk control. Without stop losses we would still be controlling our risk, but it is the risk of how much we might lose over a given time period, not on any given trade.
Now let us turn to the other function of stop losses: getting you out of losing trades . How can we do this without stop losses?
We use the opening rule to c lose trades .
The rule used in the Starter System is a trend following rule; we buy when the market has rallied, and sell when it has fallen. If a trend reverses then we would start losing money and would close our positions. But we don’t need the stop loss rule for this.
Instead we can t
rack the opening rule throughout the life of the trade and close our position when it reverses i n direction.
Stop losses ar en’t perfect
Trade-based stop loss risk control feels safer to most traders, which is why I specified them in the Starter System. There is something comforting about knowing there is a maximum amount you can lose on each trade. But it is cold comfort, as you can easil y lose more.
Here is a stop loss horror story. I bought shares in the UK
postal service, Royal Mail Group, in January 2017 for £4.47.
After some sideways movement they rallied in early 2018 reaching a high of £6.31 in May 2018, at which point my trailing stop ¹²¹
was at £4.42. Over the next few months the price drifted downwards, reaching £4.77 on 28 September. Although this was disappointing, I consoled myself that if they hit my stop at least I’d get away with a modest £0.05 loss (£4.47 versus £4.42).
Including dividends received I would even have a s mall profit.
Then late on the afternoon of Monday 1 October 2018 Royal Mail put out a profit warning, and the share price plummeted below £4.
By the time I managed to sell, the price was down to £3.60, crystallizing a loss of several thousand pounds. So much for the safety of s top losses. ¹²²
Of course, time-based risk control is not perfect. It assumes we can predict the volatility of prices with unnatural foresight.
Actually, standard deviation is the one of the most predictable
values in financial markets, although of course it can never be perfectly forecasted.
But time-based risk control only tells you what your average loss will be. It does not promise to limit your daily losses to some maximum level, like a stop loss does. Unlike stop losses time-based risk control doesn’t make promises it ca n’t deliver.
Continuous trading
With no stop loss rule the trading process look s like this: Day one: starting with no position, we check our o pening rule.
We enter into a position in the direction of the trend found by the o pening rule.
We continue to hold the position whilst the opening rule is in the sam e direction.
Leveraged Trading: A professional approach to trading FX, stocks on margin, CFDs, spread bets and futures for all traders Page 19