Professional Documents
Culture Documents
A submission to the Market Technicians Association (MTA) for the 2012 Charles H Dow Award.
Introduction
Psychology is the key to understanding the predictable and repeatable components of human behavior. This fact has long been recognized by students of market action, particularly those schooled in the discipline of technical analysis. However, the discipline of economics has long resisted the notion that man is anything but a rational calculating animal. According to the Efficient Market Hypothesis (EMH), there can be no possibility of a risk-adjusted return premium using only price data. For decades, this has led the academic world of financial theorists and the practical world of financial traders to lead separate lives in seemingly parallel universes. In one corner, the practical trader has employed charts, indicators and sundry forecasting devices to pursue profits based on publicly reported transactional data. In another corner, the academic community constructed proofs of the impossibility of such devices ever leading to profitable risk-adjusted returns. Of course, there were some brave academic souls, the pioneers of behavioral finance, who steadily and diligently amassed evidence that all was not well with the EMH. Among these, Jegadeesh and Titman have demonstrated economic profits from the following of price momentum, both relative and absolute [1]. Others, such as the Khaneman and Tversky [2], provided evidence, through their experiments that led to the Nobel prize winning Prospect Theory, that utility theory is a poor model for the process of human decision making. Academics have thus, by stages, warmed to the view that technical analysts have long held to be central to a practical understanding of market action. Recently, Kirkpatrick and Dahlquist have done much to document the modern rapprochement between academic researchers and practically-minded traders [3]. Passing now from academia and the technical analyst community, we must consider the man on the street. Recent market events have caused much despair.
If Man is always fully aware of the likely future consequences of his present actions, then how, one may ask, did financial markets manage to get themselves into such a pretty predicament? In the past five years, we have had seemingly endless market crises. It would be just fine and dandy to assume that the events of the 2008-2009 bear market, and the global credit crisis, were simply consequences of unforeseeable circumstance, the mere slings and arrows of outrageous fortune with no visible proximate cause. The charitable analyst might take this view. Picture the hapless Dinosaur of Yore who is bewitched by the fiery approach of an asteroid. How might that reptile ponder the impending demise of his species? It was an entirely unpredictable Act of God! Was this true of the Global Financial Crisis? How could sentiment be so positive in one moment, and so dire the next? This shall be the focus of our enquiry. What are the Fundamentals of Sentiment?
Cost-Basis Theory
Any professional trader who enters the market must have a theory. However devised, constructed, borrowed, or purloined, a theory is necessary to support prudent action. The market is too deceptive to merely trust a single price tick, daily bar, or reported piece of official data and consequent analyst opinion. Markets are deceitful in the most delicious and confounding fashion. It is part of their endless charm. They will Zig just as soon as Zag, especially if our hapless trading Pilgrim has placed his stop-loss limit order too close to a visible level and is mauled by the Pit Hounds of Hell. Historically, economists have assumed that absolute price levels dont matter. Traders know differently, since they are aware of the institutional constraints upon them. Where, in the whole crazy grand scheme of things, should we look first? Among all options, it is nominal debt contracts that figure prominently when gearing is involved. If you owe more than the asset is worth, rapid liquidation, even at fire-sale prices, seems rational. This psychological insight is the essence of cost-basis theory. 2
Very clearly, a trader who operates on margin will worry some if his stock should fall below its book value of purchase. He will worry greatly when overextended with an amount of borrowed money. Each hopeful uptick in price restores a sliver of equity. Each relentless downdraft in price washes away the very anchors of his soul. Hence we are bold enough to make an assertion of principle. Contrary to the clear and present assumptions of orthodox economics, we maintain that price is the fundamental variable that determines the solvency, and thus the sentiment, of geared investors.
18 16
Price
14 12 10 8 28-Jan-10
28-Apr-10
28-Jul-10
28-Oct-10
28-Jan-11
Time
The past price and volume of all trades, the tape as we call it, contains some valuable information. When shown conventionally, as per Figure 1, we see an erratic path of prices through time, with associated fluctuating trading volumes. For completeness, we have shown classical support and resistance levels, in red and green, drawn at prices where there has been significant trading congestion. However, let us now think differently and remove time from the picture. While we cannot be certain exactly what percentage of the current shareholders will still hold positions at previously traded prices, we can still deduce some principles. Where there has been very great volume traded, such as a very large block traded there is a higher probability of significant holdings still held at that cost-basis. Furthermore, if the volume was very recent then only the subsequent trading can have changed the holdings of the past. Since every share traded will change its cost-basis, it will be the newer trades that dominate over the older trades. Pulling this together, we could make a map of volume traded at price, as per Figure 2.
6.0%
11.20
5.0% 4.0% 3.0% 2.0% 1.0% 0.0% 8.00
% Volume
8.90
10.00
12.00
14.00
16.00
18.00
Price
Against this picture of history, with time removed, we can be quite certain of where the peaks lie. The only point of doubt is their relative height. Past prices and volumes do matter because they impact the market memory. The annotated peaks happen to coincide with the prior support and resistance lines. Through the lens of cost-basis theory, we see that this is a very strong mathematical property of financial markets. The history of price and volume data determines the statistical shape of the reference prices at which current investors hold securities. We can do no better than quote Jiler [4]:
Have you ever bought a stock, watched it decline in price, and yearned to sell out for what you paid? Have you ever sold a stock, watched it go up after you had sold it, and wished you had the opportunity to buy it again? Well you are not alone. These are common human reactions, and they show up on the stock charts by creating support and resistance.
The great advance of the last ten years [5, 6, 7] has been to realize that the foregoing ideas can be translated into a simple estimate of the average cost-basis of the market, and thus a simple estimate of the support and resistance levels for a market index.
It takes the form of an Exponentially Weighted Moving Average (EWMA): VWAMAT = VWAMAT-1 + TURNOVER VELOCITYT * (PRICET - VWAMAT) The formula is conventional except for one small detail. The smoothing parameter, marked as turnover velocity, is changeable and reflects the pace of market trading. It is driven by the rate-of-trade compared with market value: TURNOVER VELOCITYT = TRADED VALUET /MARKET VALUET. One simply looks up the daily traded value and divides this by the index market cap. The full justification for this formula is described in [5, 6, 7]. Adaptive moving averages have been introduced previously, notably the KAMA of Kaufman [8], and others such as the MESA and FAMA of Ehlers [9]. Each of these has been shown to have merit in certain trading situations. Our interest in the Value-Weighted Adaptive Moving Average (VWAMA) lies in the possibility of testing some of the precepts of cost-basis theory [7]. In prior work, Fries [5] noted the correlation between cost-basis and price momentum, while Grinblatt and Han [6] have also demonstrated an intimate relationship between this indicator, the disposition effect, and the persistence of price momentum. Our goal is related, but to a different purpose: 1. We seek a market signal for aggregate investor sentiment 2. We wish to investigate the trading efficacy of the moving average signal The first point will already be clear given the previous motivation. We have argued at length, that cost-basis is an important determinant of sentiment since it fixes the mark-to-market profit and loss of all investors.
Finally, we have introduced, from the above prior studies, a remarkable and perhaps unexpected connection. Exponential moving averages, of this special type, have a direct relationship to the unrealized profit-and-loss of existing holdings. The simple ratio: UNREALISED PROFIT AND LOSST = PRICET /VWAMAT 1 measures trader gains and losses, and thus investor sentiment. In words, if the current price is $12 and the average cost basis is $10, then the average unrealized profit and loss is $2. This would be a condition of positive sentiment. On the other hand, if the price is $10 and average cost basis is $12, we have a $2 loss.
In words, we seed the series with the first index price, a price 10% higher and a price that is 10% lower. Then we see how many years lapse before they converge. In general, this period is about five years. The worst of markets studied was Japan, and one can see the demonstrated pathway towards convergence after ten years in Figure 3.
Since our data set is some twenty years long, we were satisfied to proceed with the study using unedited values at the central estimate of the initial starting price.
Trading Studies
Since we have 45 markets, with daily data over 20 or more years, there are excellent grounds to embark upon some tests of simple trading rules. However, there is very little experience with testing Value Weighted Adaptive Moving Averages, so we have little to guide us but the cost-basis theory elaborated earlier. Rather than simply mine the data for all possible rule combinations, we have chosen to test only two strategies [12]. The discipline we follow is the logic of our cost-basis theory and to assume that the cost-basis level is significant to sentiment. Firstly, we examined the time-series of Profit & Loss Signal.
One can see, from Figure 4, the typical pattern of Bull and Bear trading. There are long upward trends lasting several years where the market finds support at cost-basis. These are punctuated by shorter periods of much more intense downward moves. 10
To visualize this we plot the ratio of the market index to the cost-basis index. This variable has the direct interpretation of measuring average unrealized trader gains and losses.
The corresponding chart for the United States is shown in Figure 5. Simple reasoning suggests that this quantity should be mean reverting. We can also perceive some possible evidence of a market tendency to avoid trading near cost basis. It is as though this price level repels the market from both above and below, much as the classical theory of support and resistance contends [3]. Detailed summary statistics across all 45 markets are displayed in Table 2 confirm that returns are higher in positive P&L periods and risks are lower. Following these observations, we test what we call Simple MOMA (SMOMA) based on the simple-minded idea of being short in bear markets and long in bull markets. This is a pure Stop and Reverse (SAR) system [8], with no free parameters. 11
When the index price is above the estimated cost-basis level the strategy is long. When it makes the cross-over to negative territory the strategy is short. We model trading cost as 50bps per transaction, which is 100bps per reversal in the SAR system. Since ETF or index futures are available, this seems conservative. Results for SMOMA are reported in Table 3. There are very many small losing trades and a few large winners associated with being long in the bull market phase. The second strategy is less obvious, and does have two free parameters which we fix on a somewhat arbitrary, but informed basis. Typical moving average systems employ a fast and a short moving average [8]. The speed of the MOMA is driven by market turnover, which is a market parameter. However, the actual profit experience of traders is driven by their gearing level. For simplicity, we assume that traders come in two basic levered classes. Margin traders we assume are 2:1 geared. Pattern day traders we assume are 5:1 geared. These numbers reflect the Federal Reserve System limits as a guide to the critical level when the geared trader tips into forced-sale territory. Following this idea, we constructed a classical fast-slow moving average system after the pattern of Kaufman [8] and other readily available guides for the private trader. The gearing adjustment works through the trading turnover as: A_MOMAT = A_MOMAT-1 + A * TURNOVER VELOCITYT * (INDEX PRICET A_MOMAT) The new variable, A, is an acceleration factor to the velocity which speeds up MOMA. It represents an arithmetically compounded estimate of unrealized profit and loss for a trader who is deploying day-trading capital on an A:1 geared basis. Effectively, the A factor is the amount by which one day of profit (or loss) is compounded by gearing. In this system, which is again a pure Stop and Reverse (SAR) system, we simply sell when the 5:1 accelerated MOMA (A=5) crosses the 2:1 accelerated MOMA (A=2). The strategy sounds complex, so it is better to simply visualize in chart form.
12
Figure 6 AMOMA Strategy for Australia involving x5 speed MOMA crossing x2 speed MOMA
In Figure 6, we can see the underlying BMI Index for Australia, with slow MOMA acting as support in bull phases. For our AMOMA strategy, the rule makes money versus buy and hold when it is short, and correct. Using the x5 accelerated average; the rule is short whenever it falls through the x2 accelerated average. These are covered on the reversal move, producing the sharp spike in strategy profit during the bear market periods. The rule works to capture a developing cascade of liquidation. In Table 4, we display the geometric return on the trading book, with a discount of 100bps applied to each reversal of position.
Simulation Results
Table 3 and Table 4 display the results of our simulations across all 45 markets. Less than 1/3 of the cases lead to profits after transaction costs.
13
This may be somewhat disheartening, until it is noted that the max gain to loss ratio and win versus loss trade statistics show the typical footprint of a trend following strategy. With attention to stop-losses on short sales, it seems plausible to develop a workable trading system. The key takeaway for further work is the importance of the short side since the system is benchmarked to a buy and hold strategy.
Conclusion
Our goal in this study was to explore the fundamentals of sentiment through the notion of the average cost basis of the market. We have developed what appears to be the first systematic multi-market study of simple trading strategies based on the traditional ideas of moving average crossings [8]. While this study has entered new territory, the results are encouraging of further research directed at developing a trend-following trading system that can capture and protect short-sale profits in bear markets.
14
Table 1 Local Currency Returns and Date Ranges for S&P/Citigroup Broad Market Indices
Country Days Argentina 3842 Australia 5870 Austria 5870 Belgium 5870 Brazil 4436 Canada 5870 Chile 4436 China 4435 Czech Republic 4436 Denmark 5870 Egypt 3912 Finland 5870 France 5870 Germany 5870 Greece 4436 Hong Kong 5870 Hungary 3913 India 4435 Indonesia 4436 Ireland 5870 Israel 4435 Italy 5870 Japan 5870 Malaysia 5870 Mexico 4436 Morocco 3912 Netherlands 5870 New Zealand 5870 Norway 5870 Peru 4436 Philippines 4436 Poland 4436 Portugal 4436 Russia 3913 Singapore 5870 South Africa 4436 South Korea 4436 Spain 5870 Sweden 5870 Switzerland 5870 Taiwan 4436 Thailand 4436 Turkey 4436 United Kingdom 5870 United States 5870 FirstDate 29-Dec-94 30-Jun-89 30-Jun-89 30-Jun-89 29-Dec-94 30-Jun-89 29-Dec-94 30-Dec-94 29-Dec-94 30-Jun-89 31-Dec-96 30-Jun-89 30-Jun-89 30-Jun-89 29-Dec-94 30-Jun-89 31-Dec-96 30-Dec-94 29-Dec-94 30-Jun-89 30-Dec-94 30-Jun-89 30-Jun-89 30-Jun-89 29-Dec-94 31-Dec-96 30-Jun-89 30-Jun-89 30-Jun-89 29-Dec-94 29-Dec-94 29-Dec-94 29-Dec-94 31-Dec-96 30-Jun-89 29-Dec-94 29-Dec-94 30-Jun-89 30-Jun-89 30-Jun-89 29-Dec-94 29-Dec-94 29-Dec-94 30-Jun-89 30-Jun-89 LastDate FirstValue LastValue Buy & Hold Return Annualised Return 18-Sep-09 67.28 424.46 630.91% 12.84% 30-Dec-11 93.42 301.48 322.71% 5.16% 30-Dec-11 101.14 174.98 173.02% 2.38% 30-Dec-11 116.39 197.53 169.72% 2.30% 30-Dec-11 55.23 627.58 1136.33% 14.81% 30-Dec-11 112.36 379.44 337.69% 5.36% 30-Dec-11 107.39 358.08 333.43% 7.08% 30-Dec-11 108.45 312.39 288.04% 6.20% 30-Dec-11 109.05 251.98 231.06% 4.87% 30-Dec-11 130.57 625.70 479.20% 6.96% 30-Dec-11 78.08 416.12 532.95% 11.38% 30-Dec-11 177.25 744.14 419.84% 6.35% 30-Dec-11 98.83 218.62 221.21% 3.47% 30-Dec-11 104.34 271.36 260.06% 4.19% 30-Dec-11 49.78 28.43 57.12% -3.13% 30-Dec-11 42.28 338.85 801.48% 9.35% 30-Dec-11 45.33 164.85 363.66% 8.67% 30-Dec-11 101.57 448.85 441.90% 8.81% 30-Dec-11 121.15 791.59 653.40% 11.25% 30-Dec-11 114.62 242.45 211.52% 3.27% 30-Dec-11 58.97 197.58 335.03% 7.11% 30-Dec-11 144.01 158.58 110.11% 0.41% 30-Dec-11 186.81 56.95 30.48% -4.97% 30-Dec-11 74.70 233.13 312.09% 5.01% 30-Dec-11 47.53 592.83 1247.15% 15.41% 30-Dec-11 65.93 243.83 369.82% 8.79% 30-Dec-11 97.07 263.56 271.50% 4.38% 30-Dec-11 138.68 137.89 99.44% -0.02% 30-Dec-11 143.04 421.27 294.51% 4.75% 30-Dec-11 67.05 958.12 1428.97% 16.31% 30-Dec-11 157.92 237.89 150.64% 2.35% 30-Dec-11 51.39 173.96 338.51% 7.17% 30-Dec-11 46.28 62.37 134.78% 1.71% 30-Dec-11 49.38 1323.73 2680.83% 23.59% 30-Dec-11 90.24 149.12 165.26% 2.18% 30-Dec-11 98.01 528.95 539.72% 10.05% 30-Dec-11 228.47 661.21 289.41% 6.22% 30-Dec-11 140.75 411.29 292.21% 4.71% 30-Dec-11 102.47 604.19 589.61% 7.91% 30-Dec-11 87.55 321.23 366.93% 5.74% 30-Dec-11 89.46 70.67 78.99% -1.33% 30-Dec-11 357.44 232.06 64.93% -2.42% 30-Dec-11 8.18 1257.61 15380.03% 33.12% 30-Dec-11 80.66 213.54 264.73% 4.27% 30-Dec-11 70.99 325.37 458.36% 6.75%
15
Table 2 Daily Returns: Bullish Positive P&L (Up) vs Bearish Negative P&L (Dn) plus Annualized Spread
Country DayUp RtnUp StdUp RngUp DayDn Argentina 3042 0.13% 1.89% 1.31% 799 Australia 4692 0.05% 0.80% 0.59% 1177 Austria 3270 0.07% 1.06% 0.72% 2599 Belgium 3400 0.05% 0.85% 0.60% 2469 Brazil 3741 0.10% 1.58% 1.14% 694 Canada 4071 0.05% 0.86% 0.61% 1798 Chile 3108 0.06% 0.98% 0.67% 1327 China 2739 0.14% 1.68% 1.16% 1695 Czech Republic 2248 0.11% 1.32% 0.95% 2187 Denmark 4203 0.06% 0.89% 0.64% 1666 Egypt 2194 0.14% 1.64% 1.03% 1717 Finland 3128 0.14% 1.85% 1.24% 2741 France 3916 0.07% 0.98% 0.73% 1953 Germany 3885 0.09% 1.02% 0.75% 1984 Greece 2172 0.12% 1.65% 1.12% 2263 Hong Kong 4484 0.08% 1.26% 0.88% 1385 Hungary 2793 0.13% 1.73% 1.21% 1119 India 2778 0.14% 1.53% 1.06% 1656 Indonesia 2958 0.12% 1.62% 1.08% 1477 Ireland 3094 0.05% 1.15% 0.80% 1339 Israel 3519 0.06% 1.31% 0.87% 915 Italy 3129 0.08% 1.14% 0.82% 2740 Japan 1450 0.08% 1.00% 0.72% 4419 Malaysia 3895 0.06% 1.06% 0.69% 1974 Mexico 4118 0.09% 1.44% 1.01% 317 Morocco 3625 0.04% 0.86% 0.56% 286 Netherlands 4110 0.07% 0.94% 0.67% 1759 New Zealand 3227 0.03% 0.91% 0.60% 2642 Norway 3752 0.08% 1.08% 0.80% 2117 Peru 4078 0.08% 1.53% 1.04% 357 Philippines 2008 0.07% 1.18% 0.84% 2427 Poland 3028 0.11% 1.55% 1.11% 1407 Portugal 2215 0.08% 0.98% 0.67% 2220 Russia 3412 0.15% 2.30% 1.55% 500 Singapore 3944 0.06% 0.99% 0.72% 1925 South Africa 3815 0.06% 1.13% 0.81% 620 South Korea 2841 0.09% 1.69% 1.18% 1594 Spain 3182 0.10% 1.03% 0.76% 2687 Sweden 4019 0.10% 1.20% 0.88% 1850 Switzerland 3836 0.07% 0.86% 0.63% 1902 Taiwan 2302 0.14% 1.27% 0.90% 2133 Thailand 2168 0.10% 1.28% 0.92% 2267 Turkey 3217 0.27% 2.36% 1.66% 1218 United Kingdom 4404 0.06% 0.80% 0.59% 1465 United States 4661 0.07% 0.89% 0.63% 1208 RtnDn StdDn RngDn BullPercent UpVsDn AnnUpStd AnnDnStd -0.15% 2.90% 1.96% 79% 71% 30% 46% -0.08% 1.43% 1.03% 80% 33% 13% 23% -0.05% 1.49% 0.95% 56% 32% 17% 24% -0.03% 1.42% 0.92% 58% 18% 13% 22% -0.08% 3.54% 2.25% 84% 45% 25% 56% -0.04% 1.30% 0.80% 69% 24% 14% 21% -0.03% 1.02% 0.69% 70% 22% 16% 16% -0.12% 2.40% 1.64% 62% 66% 27% 38% -0.05% 1.77% 1.18% 51% 41% 21% 28% -0.04% 1.47% 0.99% 72% 25% 14% 23% -0.06% 1.83% 1.09% 56% 50% 26% 29% -0.07% 2.04% 1.42% 53% 51% 29% 32% -0.07% 1.74% 1.22% 67% 35% 16% 28% -0.11% 1.88% 1.32% 66% 50% 16% 30% -0.10% 1.90% 1.32% 49% 55% 26% 30% -0.06% 2.06% 1.38% 76% 35% 20% 33% -0.13% 2.48% 1.71% 71% 65% 27% 39% -0.11% 1.81% 1.24% 63% 63% 24% 29% -0.05% 2.44% 1.60% 67% 42% 26% 39% -0.06% 1.93% 1.31% 70% 27% 18% 31% -0.07% 1.53% 1.06% 79% 33% 21% 24% -0.07% 1.61% 1.14% 53% 40% 18% 25% -0.04% 1.40% 0.97% 25% 31% 16% 22% -0.03% 1.94% 1.13% 66% 22% 17% 31% -0.21% 2.53% 1.83% 93% 75% 23% 40% 0.00% 0.65% 0.44% 93% 11% 14% 10% -0.09% 1.73% 1.22% 70% 41% 15% 27% -0.03% 1.10% 0.79% 55% 16% 14% 17% -0.06% 1.88% 1.29% 64% 35% 17% 30% -0.05% 1.91% 1.25% 92% 34% 24% 30% -0.02% 1.68% 1.12% 45% 23% 19% 27% -0.09% 2.00% 1.45% 68% 50% 25% 32% -0.05% 1.25% 0.84% 50% 34% 16% 20% -0.06% 5.02% 3.41% 87% 52% 36% 80% -0.07% 1.88% 1.29% 67% 32% 16% 30% -0.08% 1.63% 1.09% 86% 35% 18% 26% -0.03% 2.54% 1.72% 64% 29% 27% 40% -0.06% 1.65% 1.15% 54% 39% 16% 26% -0.08% 2.01% 1.43% 68% 46% 19% 32% -0.05% 1.51% 1.05% 67% 30% 14% 24% -0.13% 1.86% 1.35% 52% 68% 20% 29% -0.08% 2.41% 1.63% 49% 43% 20% 38% -0.19% 3.15% 2.17% 73% 116% 37% 50% -0.10% 1.60% 1.14% 75% 40% 13% 25% -0.10% 1.85% 1.29% 79% 40% 14% 29%
16
17
18
References
1. Jegadeesh, Narishman and Sheridan Titman, Returns to Buying Winners
Complete Resource for Financial Market Technicians 2nd Edn, New Jersey, NJ FT Press (2005)
4. Jiler, William, How Charts Can Help You in the Stock Market, New York, NY
Investor Sentiment, Credit Suisse First Boston, Quantitative Research Note, Sydney August (2002).
8. Kaufman, Perry J, New Trading Systems and Methods 4th Edn, New York,
19
20