A Better Trend – Kaufman Signals

Are we bored yet? We may be tired of trend following, but we should never ignore it just because it’s boring.

Consider these points:

Macrotrend systems have had the best track record of any method for the past 30+ years.
They are based on a sound premise – following the reaction to economic policy.
They don’t trade often (which bothers the brokerage houses).
They work across nearly every market.
They work across a large range of time frames (calculation periods).
They capture the “fat tail” of price moves, the really big moves.
Therefore, they are robust. Very few trading strategies are robust.
And, yes, they can be boring.

By robustness we mean

We apply the same rules and parameters to all markets
“Loose pants fit everyone” means that some markets won’t trade as well as others when you use the same parameters and rules, but fine-tuning is a guarantee of failure.

There are also these advantages

Knowing the trend allows you to filter trading signals from other strategies.
Knowing the trend allows you to weight trades with better expectations.
Knowing the trend helps with options strategies.
The concept is transparent, not a “black box,” that is, we understand what it’s doing.

Overall, knowing the trend of prices can improve any trading. But, what’s the downside?

They are slow getting out of a trade, often giving up a large part of recent gains.
They are insensitive to current events and often seem stupid.
They require discipline to follow.
They can be frustrating.

Those are the pros and cons. For my part, I see mostly pros. It is constantly a surprise how anything so straight-forward can work so well.

A Sound Premise

An overwhelming consideration in using a trading strategy is whether it has a sound premise. That distinguishes methods that have been “found” by testing indicators and rules on a computer, then settling on the one that was most profitable, or most reliable, or most anything. “Discovering” a system through testing is a sure way to lose money. The right way is to first determine the reason why a strategy should work, then implementing it. A macrotrend system is based on identifying the direction of prices based on government economic policy (mostly interest rates) or long-term shifts in supply and demand, or a change in the stability of a country (geopolitical risk). While there are other reasonable approaches, such as arbitrage, this is a sound premise.

All Trends Are Not Quite Alike

To get more specific, what trend method do you use? A moving average? A linear regression? An N-day breakout? Something more sophisticated? Each method has unique way of identifying the trend. A moving average lags the market, linear regression tries to forecast ahead, and a breakout takes a current high or low as the new direction.

While one is better than another for some markets some of the time, in the long-run, over many years of data and over a wide sample of markets, the results are remarkably the same. It’s the price action that makes the biggest difference. If prices trend, all trend-following methods work. If it doesn’t trend then they all lose. Welcome to reality.

The subtle difference between the various techniques is in the risk. The simple moving average has many small losses and fewer large gains. The breakout has fewer trades with large interim losses, but it has a high percentage of profitable trades when closed out. The linear regression is somewhere in between the two, not as fast as the moving average and not as slow as the breakout, given the same time horizon. You’ll want to explore this yourself to find the trend that best suits you.

Choosing the Trend Speed

Most traders end up with a single moving average that they track. But which one? It doesn’t take long to find out that the long-term trends are more reliable than short-term, that is, calculation ranges from 40 days and higher tend to be profitable for most markets. It’s not even clear that there are consistently successful short-term trends. As an example, Chart 1 shows a test of moving averages on Apple (AAPL) prices, for periods from 10 to 150 days. There is a clear tendency for more profits as the trend periods get longer. Experience and testing shows that this is generally true of most stocks and futures markets because noise interferes with faster trends.

Do you choose the most successful, the 140-day trend? That depends on whether you believe that past performance predicts future results. I do, for this method, but there is considerable variability from year to year. We can reduce that risk by choosing more than one calculation period in the same way we diversify a portfolio by choosing more than one stock. Then we’ll get a somewhat lower than “perfect” result each year, but expect more stability. As we choose more combinations our results will get closer to the average. Statistics says that after choosing four periods, all further results from diversification diminish quickly.

How do you trade more than one trend at the same time? Chart 2 shows the EuroSchatz futures with a 20, 40, and 80-day moving average. When all three are trending up we would be fully long, and when one is trending down and the other two up, we would be long 1/3. We net out the long and short trends. In Chart 3 we see the same trends for Apple. Most macrotrends would start at 40-days and go to periods of 120 to 200, but in these examples, using 20 days shows the differences clearly.

Seeking an Average Performance

Choosing a single trend means that you expect this trend to outperform the average. Given a choice of all trend periods between 20 and 200, what are your chances of choosing the one the will perform best next year? It would be the same as choosing a single market to trade instead of a diversified portfolio. Could you have figured out that Apple would have gone to $800? That gold would go to $1800, or crude oil to $150? We diversify to improve our chances of being right, knowing that we give up any chance of an extreme profit. It’s a good trade-off.

The same is true for choosing calculation periods. Long-term trend following is unique because it is profitable (to different degrees) over a wide range of time frames, as seen in the Apple example (Chart 1). If 70% of all trends are profitable over a wide range of periods, then why not trade a fair sample of them and target the average? While you can’t get the highest profit, you also can’t get the worst loss. Instead you get stability.

A Note About Test Distribution

At the risk of being too mathematical, the tests in Chart 1 misrepresent the results. Technically, they are correct, but they make it appear as though the larger calculation periods are more stable than they really are. That’s because the difference between a 10-day average and a 20-day average is 100% (the 20 days has twice the number of prices as the 10 day), while the difference between the 190 and 200-day averages are only about 5%. So we should expect only a slight difference in the results of the 190 and 200 day calculations. That will make the right end of the chart look more stable. To be more accurate, each test should be incremented by the same percentage, for example 20-40-80-160, or 20-30-45 and so on. Most test platforms don’t allow that, but it’s good to be aware of the problem.

Averaging Into a Trade

We can apply the same idea of averaging the trend positions to the way we enter and exit a trade. We could execute the entire position at once, on the open of the next day, or we can divide the position size into equal parts and execute over a span of days. For example, for a trade of 100 shares we could execute 50 on the first day and 50 on the second, third or fourth day. Or, execute in 4 equal parts, each 2, 3, or 4 days apart. We then get an average price rather than counting on “all-or-nothing.”

Averaging-in works best for long-term trends because they are not sensitive to a price move on a specific day. If you consider an 80-day average, today’s price move is 1/80^th of the total. A new buy signals will be the aggregate of many days and today’s change may be up, but tomorrow and the following day may be down. An average entry price is always safe.

Charts 4, 5, and 6 show the results of testing the EuroSchatz from 1990 through 2013 as “heat maps.” The EuroSchatz is the European equivalent of the U.S. 2-year note, but serves as a good example here (there are many good examples). In Chart 4 and 5 we’ve tested moving average periods from 20 to 120 days (in the left column), and entries in 3 parts separated by 1 day, 2, days, up to 5. The column marked “0” enters the entire trade on the first day. Chart 4 gives the net profit/loss and Chart 5 the profit factor (total profits divided by total losses), a reasonable measure of reward to risk.

Chart 4, profits, shows a successful band through the center of the table at 60-70. While all the cells are profitable, column 1 varies from dark red to green, showing some instability. The farthest right column, although not the most profitable, seems to be the most stable. The picture is much clearer in Chart 5, the profit-factor. Here the worse results are clearly towards the upper left and the best towards the lower right. I have always judged success by a risk/reward ratio rather than only profits.

In Chart 6 we’ve copied the calculation periods separated by the same percentage of days, showing the same results as in Chart 5.

These are not isolated results, but generally true. Not all examples are quite as clear, but nearly all show that averaging into a trade is a better strategy for long-term trend following than entering the entire position on any one day. And, it improves liquidity for large hedge funds that often use this strategy.

One particular advantage of averaging-in is when the trend changes direction sooner than expected, what we all call a “false signal.” If we are entering and exiting in three parts each, then there are six trades needed to reverse a position. During that interval, if the trend reverses, then the entire position was not exited or entered, resulting in a substantial savings.

Scaling In and Out with Different Position Sizes

We calculate the size of the trade for stocks by dividing the amount to be traded, say $10,000, by the current price. While not at all perfect, it does attempt to risk-adjust the position size. When there are five different calculation periods and three parts for averaging into a trade, there are a total of 15 buy orders to reach a full long position. Then each order gets 1/15^th of the investment, consistent with the idea of equally weighting.

However, between the first and the last order, the price can change. We want to keep our position tuned to the current price level, so that we reduce our size as prices move higher and increase as they move lower. When a new leg is to be entered, we calculated the total new position size based on today’s price, then add or subtract whatever is necessary to have a total position size consistent with the current price level. For example, we are investing $10,000 in Bank of America (BAC), therefore, each leg gets 1/15^th, or $666.

On our first buy signal, the price is $10, therefore we buy 66 shares.
At our second buy signal, the price has dropped to $9, so we want a total of $1332 at $9 = 148 shares. Had we only considered the 1/15 allocation, we would add 74 shares, but now we add 148 – 66 = 82 shares.
On the third signal the price is $15, therefore, we should be holding 3/15ths or $2000 of shares. At $15 we get 133, larger than the 148 we are currently holding even though we have 3 parts of our total position, rather than 2. We sell 15 shares to bring the risk into line.

This process allows us to continually adjust the risk of our position to the current risk of the market without unnecessary costs. If the adjustment was too small, we would simply wait for the next order.

The Switching Threshold

There is a point where we could be trading too many combinations. While the more you trade, the closer you get to the average, you also need to consider the costs. With more combinations there are likely to be small changes in the net position each day, which means a trading commission each day. There are two ways to solve this: fewer combinations, or changes in net position size greater than some threshold, say 15%. If you position changes from 200 to 210 shares, then don’t do anything. Wait until the change is at least 30 shares before making an adjustment.

Execute Trades on the Next Open

There is something to be gained by executing trading signals as soon as possible, but it is not always practical. The choices are

Use the closing price to calculate signals, then execute in the aftermarket
Use prices 15 minutes before the close, execute on the close, then rerun and adjust if needed
Use the close for the signals, then execute on the next open

By far the most realistic and convenient is to use the closing prices to generate the signals, then execute on, or just after, the open of the next trading session. Given practice, most trading desks are able to execute with very little slippage. This delay has only a minor effect on long-term trends, which are not reacting to specific moves on the day of entry. A change in the long-term trend can just as easily be caused by the oldest data dropping off as the change in the new data

Summary of Features for A Better Trend

Macrotrend time frames, calculation periods from 40 to 120 days
Multiple trends, spaced by percent change in calculation period
Averaging in over n days, spaced equally apart.
Actual position is net of all trends, equally weighted
Position size determined by (average true range) ATR for futures, but the price for stocks
When new positions are added or reduced, net the total at current price or volatility
To reduce the cost of trading, don’t add or reduce your position unless it changes by 15%
No stops are used because they fight with the trend
Portfolios are long-only, although short sales are monitored

Note: A version of this article was published in the May 2014 edition of Technical Analysis of Stocks & Commodities