Do some instruments offer better returns than others?
Q: “What is the best market to learn to trade for a beginner?”
A: “The best market to learn to trade for a beginner I would say is definitely Forex. Not because it’s the most liquid market in the world that runs 24/7 and all the big bucks are here, but because in my opinion it is way easier to trade than futures or stocks.”
A: “All you need is to learn Forex trading and the rest of the markets will be like piece of cak e for you. “
A: “Don’t know a lot about futures trading. But can tell you something about Forex and Stocks. ... On Forex you can use leverage and make more money for less cost. Forex trading is faster, more risky, but may be more profitable.”
Extremely poor advice in response to a question on quora.com The point of trading is to make money. Specifically, we want to maximise the profits we can make, net of any costs we have to pay.
Like the posters quoted above, you probably assume that likely profitability is the most important factor when deciding which instruments to trade. To begin with, let’s focus on trading profits before any holding or transaction costs have been paid: pre-cost returns. Figure 1 shows the back-tested pre-cost Sharpe ratio (SR) for the Starter System, across a large number of 20 20
different instruments 74 78 which I currently trade in my own portfolio. The average SR is around 0.24, but there is a huge difference between various instruments.
Figure 1: back-tested pre-cost Sharpe ratio (SR) for the Starter System
It’s clearly a no brainer. We should all be trading Italian government bonds (BTP), or perhaps French bonds (OAT), or maybe Korean bonds (KR3 and KR10). Also, we should stay away from the Swiss and Dutch equity indices (SMI and AEX respectively). They are clearly a money loser. For what it’s worth, most Forex markets (like the instrument labeled AUD, which is AUDUSD) fall somewhere in the middle.
But this ignores something very important: statistical uncertainty . The average performance of a trading system is a statistical parameter , which we can estimate, but which cannot be kn own exactly.
Statistical uncertainty of paramet er estimates Statistical uncertainty can be a difficult concept to get your head around. However, it’s a key part of understanding how to trade properly, so it’s vital that yo u grasp it.
Imagine that I am trying to sell you a trading system. I have traded it with my own money, and I have a back-test. Your first question should be, “How long is the back-test?” I admit to you that I have only a single day of back-te sted trades.
Is one day long enough to prove that a particular trading system is worthwhile? Hopefully your instinctive answer is a firm no.
Fortunately, I have another system which I have been trading for five years. Would you trust this system more than the other? Of cours e you would.
Longer track records give us more confidence, because we intuitively understand that the performance of a trading system isn’t something that we can know precisely. With more evidence we can be more confident about the likely performance.
Statistical techniques allow us to go beyond this intuition and calculate how certain or uncertain we can be about the likely performance of a trading system given a track record or back-test. Consider a non-financial example. Imagine that I am standing outside a locked room full of people, and I am trying to work out the average height of everyone in the room. Before I arrived a certain number of people left the room, so they could be measured. The people who left the room were selected randomly, and in the jargon, they ar e my sample .
Suppose five people left the room, and I measure their heights.
If I assume that these people are selected randomly, then my best guess for the average height of everyone left in the room will be equal to the average height of my sample. This is also known as my central estimate for the average height in the room.
However, I cannot be certain that my central estimate is accurate. The people in my random sample might be unusually tall or short. I can get a more accurate estimate by asking for a larger sample, but I can never be exactly sure of the average height of those left in the room.
Now let us relate this to trading systems. Imagine there is a mysterious black box on your desk. The box contains all the possible daily returns for a trading system, each written on a piece of paper. We cannot see inside the black box, which means we can never know the real properties of the system. In particular, we can only guess at what the future return or Sharpe ratio (S R) might be.
Whenever we run a back-test the black box spits out a stack of randomly selected daily returns, through a slot. For example, for a one year back-test we get about 250 returns. These returns are my sample . We can estimate the average and standard deviation across all those returns, and when combined with the interest rate we can borrow at, they can be used to calculate the SR. The Sharpe ratio we estimate from a back-test will be our best guess for what the true SR is for all the returns left inside the box;
what we actually earn once we start trading. This is the central estimat e of the SR.
However, we do not know if the random set of back-tested returns is representative of all the returns that are left in the black box. We might have got lucky, and pulled out some unusually good returns, or especially unlucky and got a bunch of bad returns. We have a good deal of uncertainty about how good or bad the returns really are. The longer the back-test, the less uncertainty we have, but we never know for sure what is left ins ide the box.
So, we can never say exactly what the SR is inside the black box, but we can use statistical techniques to quantify how uncertain 20 20
we are. 74 79 Of course, there isn’t really a black box that spits out paper, but the same techniques can be used to calculate how much confidence we should have in back tested returns, given the length and properties of a particula r back test.
Figure 2 shows the effect of statistical uncertainty on our estimate of Sharpe ratio (SR) for different instruments. I have done this using a box and whiskers plot , which is a type of graph used to illustrate uncertainty.
Figure 2: Estimates of Sharpe ratio for various markets
In the middle of each vertical box is a horizontal line showing the average Sharpe ratio (SR) for a given instrument. These are identical to the values in figure 1. The other components of the plot show how confident we can be about these estimates of SR.
There is roughly a two-thirds chance that the true SR lies within the boxed area (about 68% to be a little more exact). The statistical jargon for a range of uncertainty like this is a confidence interval . There is also a 95% chance that the true SR
lies between the top and bottom ends of the lines extending from 20 20
the box (these are the ‘ whiskers’). 75 70
Taking Italian BTP bonds as an example, there is a two-thirds chance the SR is between 0.64 and 2.5, and a 95% chance it is between –0.26 and +3.43. Where we have a longer back-test history we can be a little more confident about our results, and the confidence intervals will be narrower. For gold, whose back-test goes back several decades, there is a slightly narrower confidence interval: a 95% probability that the SR is between –0.
46 and 0.72.
The key conclusion from figure 2 is that it is impossible to say, with any degree of certainty, that one instrument has performed better than another in their back-tested performance. It is possible that the best performing instrument (BTP) has a SR below
–0.2, and equally possible that the worst (SMI) has a SR above 1.
Hence: we should ignore pre-cost returns when deciding which instrument to trade.
Sizing and costs
If we assume that all instruments have the same expected pre-cost performance, then we should trade the instrument and leveraged product with the lowest expected costs . Sadly, there is a snag with this approach. The cheapest instruments and products tend to 20
require the largest amounts of capital. 75¹ Only very wealthy traders can use t
he very cheapest instruments. Regular people have to carefully consider what they can afford to spend on costs, given the limited cash they have available for trading.
The one-third of costs rule: A speed limi t on trading Clearly it would be madness to spend all of your expected profits on trading costs. On the other hand, we have to spend something
, otherwise we would not be trading at all. What is the correct proportion?
There is no single correct value, but my own rule is to limit my costs to one-third of my expected return . One-third might seem rather conservative, but there is considerable uncertainty about what our likely returns will be (look again at figure 2 if you don’t believe me). If your returns are half what you expected, sticking to the one-third rule means that you are left with something after paying costs. Because trading faster costs more, this maximum cost level acts as a speed limit on our tradi ng activity.
If both returns and costs are expressed in risk-adjusted terms, then this implies spending no more than one-third of our expected Sharpe rat io on costs.
Formula 11: Speed limit
Speed limit = Expected Shar pe ratio ÷ 3
The Starter System has an expected Sharpe ratio before costs of 0.24 (I’ll explain why shortly), implying that our risk-adjusted costs should not exceed 0.08.
Speed limit = 0.24 ÷ 3 = 0.08 (maximum total risk adjusted cos ts per year)
One-third is an absolute maximum, but ideally you should spend less. For my own trading I spend about 5% of my expected pre-cost returns on costs: around one-sixth of the speed limit. However, I trade relatively cheap futures which are not accessible to those with less capital. If you are an FX, CFD or spread better you
will have to spend a little more. Just don’t break the speed limit.
In table 3, I analyse the average costs and minimum capital levels for different leveraged products. Ultra-cheap futures are only available to those with relatively deep pockets. Margin trading is also quite cheap. In theory, you can start margin trading with just a few thousand dollars, although many brokers require a minimum account value of five or ten thous and dollars.
If you can trade them, spread bets have an even lower capital entry level, but you need to be careful as many are too pricey to trade economically. Spot FX is the most democratic of all leveraged products, accessible to the very smallest traders, but also relatively expensive. With around eleven thousand dollars, or the equivalent in pounds, you can access CFDs. Although they are quite pricey, around a third of the CFDs I analysed are still cheap enough to trade the Sta rter System.
Table 3: Cheaper products need more capital to trade Average minimum capital to run Starter System and average risk-adjusted trading cost, for each leveraged product type averaged across instruments.
Using current exchange rates. Cost calculations using appendix B, and minimum capital calculations using formula 21.
• Not currently available to US traders.
Why is spot FX often expensi ve to trade?
Spot FX is the most accessible market for traders without much capital, but unfortunately it is often expensive. We can either day trade spot FX, closing our positions each day, or trade more slowly as in the Starter System. Both methods can have prohibitively high costs if you choose the w rong broker.
Day trading is not cost effective for many products, with the possible exception of the very cheapest futures markets (I explain why at the end of this chapter). Spot FX is no different, despite day traders avoiding the cost of overnight funding.
Opening and closing a single trade every day in a relatively cheap market like AUDUSD would still cost around 0.35 a year in risk-adjusted terms. This is much higher than the expected Sharpe ratio of the Starter System, which comes in at 0.24 (I explain where this value comes from later).
Longer term trading in spot FX is a little cheaper, but still very expensive relative to other products. This is because the
holding costs for FX are often very high. Many brokers charge at least 2% a year as a funding spread, which is the main holding cost for FX traders. Currencies have instrument risk of between 6% a year (for relatively stable developed market pairs like GBPUSD) and 15% (for emerging markets like USDMXN). On a risk-adjusted basis a holding cost of 2% translates to 2% ÷ 6% = 0.33
for a stable developed market pair, and 2% ÷ 15% = 0.13 for emerging markets. These are well above my speed limit of 0.08
units of risk-adjusted costs per year (calculated back a few pages back).
In their advertising, FX brokers are fond of trumpeting their tight spreads and generous levels of allowable leverage. But unless you are day trading, execution spreads make up only a fraction of total costs in FX, and excessive levels of permitted leverage are a curse, not a blessing. You should be looking for an FX broker who is happy to operate with a funding spread of 0.4% or lower, depending on how volatile the market is.
The FX broker I used to calculate the costs in this book has funding spreads between 0.25% and 0.5%, making FX trading just a bout viable.
Let us return to the problem we are trying to resolve: which instrument should we trade, and with whi ch product?
The answer will depend on how much capital you have, and whether you are a UK trader, or based in the US. Table 4 shows the cheapest instruments and products for UK traders with up to
£10,000 in capital.
Table 4: Indicative costs for UK traders with minimum capital up to £10,000
Minimum capital has been rounded up. Cost calculations using appendix B, and minimum capital calculations using formula 21. To save space only instruments with risk-adjusted costs less than 0.06 per yea r are shown.
Table 5 shows the cheapest instruments and products for US
traders with up to $20,000 in capital. Without access to spread bets there are more limited options for traders with less cash in the ir accounts.
Table 5: Indicative costs for US traders with minimum capital up to $20,000
Minimum capital has been rounded up. Cost calculations using appendix B, and minimum capital calculations using formula 21.
Products with risk-adjusted costs greater than my speed limit of 0.08 per year have be en excluded.
Dated versus u ndated costs
You may have noticed that the tables above contain relatively few undated products: daily funded spread bets, and cash CFDs (margin trades and spot FX are honourable exceptions). This isn’t an accident – they are all more expensive than their dated counterparts (quarterly spread bets, futures, and products based o n futures).
Why?
Undated products tend to have narrower spreads and hence lower transaction costs, but higher holding costs in the form of interest margins charged by brokers on funding. However, these holding costs can be avoided if positions are closed each day.
Dated products have wider spreads, but lower holding costs. We expect undated products to make sense for people who trade very frequently, whilst slower traders should prefer dat ed products.
Let’s consider some actual figures. Table 6 shows the total risk-adjusted costs at different trading frequencies for three different leveraged products, all trading the same underlying instrument: gold. I’ve also broken out the holding cost on the bottom row.
For fewer than 50 trades each year the dated CFD (based on the future) is cheaper than the undated cash CFD. For greater than 50
trades annually, their positions are reversed. But this is a moot point, because at that sort of trading frequency we are spending far too much on costs, well above my speed limit of 0.08 Sharpe ratio units.
Table 6: If you trade quickly, undated products are relatively cheap – but still ve ry expensive
Risk-adjusted trading cost at different trading frequen cy for gold.
• Exceeds the speed-limit of 0.08 risk-adjusted cost per year
, formula 11
** Two trades every day, no positions hel d overnight
* Trading frequency of St arter System Cost calculations usin g ap
pendix B
Gold futures are much cheaper than both types of CFD, but the minimum capital required is around $117,000 (£88,000). This is far too high for most traders. Incidentally, spread bets would have a similar c ost to CFDs.
You may be surprised to see that the Starter System is only expected to do 5.4 trades on average each year. I discuss why I set up the system in thi s way later.
Table 7 shows the cheapest instruments and products for traders with very large amounts of capital; they are all futures.
Table 7: Wealthy traders pay lower trading costs Expected costs for UK / US traders with large amounts of capital.
Minimum capital has been rounded up. Cost calculations using appendix B, and minimum capital calculations using formula 21. To save space only the six cheapest instrument s are shown.
The figures in this chapter are all indicative and are based on current market conditions and the costs currently charged by the brokers I use. If you are serious about trading, then you should recalculate and check them yourself, using the formulas in appendix B.
Product choice and tax fo r UK traders All the cost figures I’ve shown here ignore the impact of taxes.
Tax treatment in the US is fairly similar across different leveraged products; at least the differences aren’t large enough that you should consider taxes when deciding which produ ct to trade.
However, this isn’t true on the other side of the Atlantic. UK
traders who trade shares, futures, spot FX and CFDs have to pay Capital Gains Tax (CGT) on their profits above an annual threshold. At the time of writing the threshold is £11,700, and tax is charged at a rate of 10% or 20% (if you have other income it will probably be 20%). But UK spread betters will usually 20
avoid payi ng any tax. 75²
As I explain later in this chapter, the sort of return you should expect on the Starter System is around 5% a year before costs.
Unless you have hundreds of thousands of pounds in capital, it is unlikely you will be making £11,700 or more with the Sta rter System.
Leveraged Trading: A professional approach to trading FX, stocks on margin, CFDs, spread bets and futures for all traders Page 9
