Martin Zweig Winning on Wall Street
Page 14
3/20/89
271.78
+13.1 5.2
8/24/89
307.51
+22.5 5.8
5/17/94
376.96
+12.7 13.8
l7/6/95
424.91
+13.2
8.6
3/20/96 *
$10,000 becomes: $219,564 223.8 mo. $1,746 137.1 mo.
Annualized return = +18.1% -14.1%
Buy-and-hold return = +4.6% per year
The right-hand side of the table lists all the sell signals, the first of which was on March 10, 1966. By the luck of the draw, the first signal just happens to be a sell signal, which is why you do not see a prior buy signal. There would have been a buy signal sometime back in 1965, which would have led to the sell in March of 1966. The sell signal side of the table also gives the percentage changes and the number of months held. For example, the March 10, 1966, sell signal was in effect eight months until the subsequent buy on November 16, 1966. Over that time the ZUPI declined 13.3%.
GRAPH J
Ned Davis Research
To date there have been fourteen sell signals and fourteen buy signals. At the bottom of the table you’ll see that $10,000, invested only during the buy signals periods, would have become $219,564 in a total of 224 months or roughly 18½ years. That equals a healthy return rate of 18.1% a year. That does not include interest, dividends, or commissions. Conversely, had you fought the Fed and fought the tape and unwisely bought stocks during the sell signal periods, your initial $10,000 would have shrunk to only $1,746 in a cumulative period of 137 months, or roughly 10 years. That works out to an annualized loss rate of 14.1%.
During this 30-year period, had you purchased the ZUPI and held it over the entire span without trading at all, you would have earned 4.6% a year. So, the buy signals produced an annualized return about 14 percentage points better than the buy-and-hold result, while the market during sell signal periods underperformed buy-and-hold by more than 20 percentage points a year. Those are excellent spreads against the market, especially for a relatively simple model that averages only about one trade per year.
The table shows that thirteen of the fourteen buy signals produced a profit, with the only loss a minuscule 1.6%. In eight of the fourteen cases, on the sell signals the market went down, and, in two of the six times it rose, the gains were minimal. The worst sell signal loss was a market rise of 15.2% on the sell given May 31, 1977. Note, though, that the sell stayed in effect for thirty-five months, nearly three full years. During that period interest rates were high, and had you invested your money in Treasury bills (not even the highest-yielding short-term instrument), you would have made a total return of 27.1%, nearly double the market’s capital appreciation over that time (although I have not included dividends at this point).
Most of the buy signal gains are very substantial, such as the 43.8% gain beginning in 1966; the 32.2% gain from the May 1970 signal; the 76% profit on the October 1974 buy signal; the 60.9% increase on the September 1981 signal; and the 65.3% rise on the October 1984 signal. When the big bull markets arrived, the Super Model did its job by getting you in very quickly and staying bullish throughout the bulk of the advances that followed.
On the other hand, the model also worked very well on the sell signals, managing to keep the investor out of the worst bear markets during the period. In particular, the Super Model gave a terrific sell signal on the last day of 1968, after which prices crashed 45.7% in the next seventeen months. The Super Model also performed well by giving a well-timed sell signal in June 1972, following which prices skidded 37.1% over the next year and a half or so. After a very brief foray into bullish territory, the Super Model reversed course, giving an accurate sell signal March 29, 1974. The worst part of the worst bear market since the thirties ensued in the next seven months, during which prices plunged an additional 33.3%. The Super Model anticipated “Black Monday” with a sell signal on September 18, 1987, following which prices plummeted 30.3%.
Graph J (pp. 110–111) plots the Super Model back to 1966 against the Zweig Unweighted Price Index. Crossings to above the upper dotted line correspond to the buy signals while crossings to below the lower dotted line correspond to sell signals.
Table 22 shows the same study when the Super Model is traded against the Standard & Poor’s 500 Price Index. Here twelve of the fourteen buy signals were profitable, with the loser showing a modest dip of 5.3% in 1980–81 and 3.2% in 1989. Eight of the fourteen sell signals led to market declines, with two of the five failures showing only very small advances. The largest market rally on a sell signal was 23.3% in May, 1994. At the bottom of table 22 you’ll see that the annualized return on the buy signals when traded against the S&P 500 is 14.5% versus merely 6.8% for buy-and-hold. Conversely, the S&P declined by 4.5% per year during sell signal periods, or over 13 percentage points worse than buy-and-hold.
TABLE 22
SUPER MODEL VS. STANDARD & POOR’S 500 INDEX: 1966 to 1996
BUY SIGNALS SELL SIGNALS
Date S&P % Change No. of Months Date S&P % Change No. of Months
3/10/66
88.96
-7.4 8.0
11/16/66
82.37
+19.4 20.5
7/26/68
98.34
+.5 2.0
8/30/68
98.86
+5.1 4.0
12/31/68
103.86
-26.3 17.0
5/29/70
76.55
+29.5 13.5
7/16/71
99.11
-7.6 3.0
11/19/71
91.61
+17.3 7.5
6/26/72
107.48
-15.4 19.5
2/14/74
90.95
+3.3 1.5
3/29/74
93.98
-25.4 7.0
10/25/74
70.12
+37.1 31.0
5/31/77
96.12
+10.6 35.0
5/6/80
106.25
+21.6 7.0
12/12/80
129.23
+5.7 .5
12/26/80
136.57
-5.3 6.5
7/10/81
129.37
-9.4 2.5
9/21/81
117.24
+41.5 25.0
10/21/83
165.95
-.1 12.0
10/15/84
165.77
+89.9 35.5
9/18/87
314.86
-20.9 3.0
12/18/87
249.16
+1.5 5.0
5/20/88
253.02
+9.4 8.6
2/7/89
299.63
-3.2 1.3
3/20/89
289.92
+21.2 5.2
8/24/89
351.52
+27.8 58.0
5/17/94
499.37
+23.3 13.8
7/6/95
553.99
+17.3
8.6
3/20/96 *
$10,000 becomes: $123,196 223.8 mo. $5,930 137.10 mo.
Annualized return = +14.5% -4.5%
Buy-and-hold return = +6.8% per year
How would you have fared had you used the Super Model since 1966? The question is answered in table 23, which shows the returns made by buying the Zweig Unweighted Price Index, or baskets of stocks roughly equivalent to it, when trading is guided by the Super Model. The approach assumes buying the ZUPI or its equivalent on the buy signals and selling it and going 100% into Treasury bills on the sell signals. Dividends are included. The first two columns list the dates and types of signal. On sell signals, the next three columns are deliberately left blank because they are irrelevant. How the stock market actually did during the sell signals was shown earlier anyhow, in tables 21 and 22. The next-to-last column on
the right shows the interest you would have earned on the Treasury bills while staying out of the stock market during the sell signal periods. The right-hand column shows what would have happened to $10,000 when following the Super Model. For example, upon the sell signal of March 10, 1966, you would have gone into Treasury bills and earned 3.2% over the next eight months, until a buy signal was given on November 16, 1966. That meant that the initial stake of $10,000 would have grown to $10,320 in that time, as seen in the top entry of the right-hand column.
Next, the second line of table 23 shows what happened after the buy signal of November 16, 1966. Column 3 indicates that the ZUPI appreciated 43.8% over the next twenty months or so before the sell signal of July 26, 1968. During that span it is estimated that dividends earned on the stocks amounted to about 4.3%, as seen in column 4. Column 5 shows the total return on buy signals, which is the sum of the appreciation in column 3 plus the estimated dividends in column 4, in this case a nice 48.1%. When that return is earned on the $10,320 at the beginning of the signal, it brings the portfolio value up to $15,289. All the signals then follow through the remainder of the table.
Note in column 5 that the total return on the buy signals was profitable in virtually every case. The interest earned on Treasury bills on the sell signals in column 6 was always profitable. So over the twenty-seven years during the test period, the investor was earning money, except for a brief period in 1989. Of course, there would have been short-run spans within a buy signal period where the market might have gone down for a while, but by the end of the signal the market was always higher, and dividends added another kicker.
TABLE 23
TOTAL RETURNS ON SUPER MODEL VS. ZWEIG UNWEIGHTED PRICE INDEX: 1966 to 1996
Date
Signal
% Appreiation on Buys
Estimated Dividents on Buys
Total Return on Buys
Internet Earned in Sell Periods
$10,000 Growth
3/10/66
Sell +3.2 $10,320
11/16/66
Buy +43.8 +4.3 +48.1 15,289
7/26/68
Sell +.9 15,421
8/30/68
Buy +9.4 +.7 +10.1 16,979
12/31/68
Sell +9.8 18,643
5/29/70
Buy +32.2 +3.7 +35.9 25,336
7/16/71
Sell +1.4 25,690
11/19/71
Buy +11.2 +1.5 +12.7 28,953
6/26/72
Sell +10.0 31,849
2/14/74
Buy +2.4 +.3 +2.7 32,708
3/29/74
Sell +4.2 34,082
10/15/74
Buy +76.0 +11.2 +87.2 63,802
5/31/77
Sell +27.1 81,092
5/6/80
Buy +21.6 +2.6 +24.2 100,717
12/12/80
Sell +.7 101,422
12/26/80
Buy +6.8 +1.9 +8.7 110,245
7/10/81
Sell +3.1 113,663
9/21/81
Buy +60.9 +9.8 +70.7 194,022
10/21/83
Sell +9.2 211,872
10/15/84
Buy +65.3 +7.1 +72.4 365,267
9/18/87
Sell +1.6 371,112
12/18/87
Buy +10.9 +1.0 +11.9 415,274
5/20/88
Sell +3.6 430,224
2/7/89
Buy -1.6 +.3 -1.3 424,631
3/20/89
Sell +3.5 439,493
8/24/89
Buy +22.5 +12.4 +34.9 592,876
5/17/94
Sell +5.87 627,678
7/6/95
Buy
+13.2
+1.5
+14.7
719,947
Total Super Model return: $719,947
Annualized return on Super Model: +15.3%
Buy-and-hold on S&P 500 (including dividends): $85,474
Annualized return on buy-and-hold: +7.4%
At the bottom of the table you’ll see that the total value of the portfolio grew from $10,000 to $719,947 in just over twenty-nine years. That equals an annualized return of 15.3%. Alternatively, had you purchased the ZUPI and held it twenty-seven years and received and reinvested all dividends throughout that period, $10,000 would have grown to only $85,474, an annualized return of just 7.4%. Thus, trading with the model produced a per annum gain more than double that of buy-and-hold. Furthermore, it did so while being out of the market and at zero risk over one-third of the time. So, on a risk-adjusted basis, the returns are even better since you would have earned 2.3 times the annualized return but taken on only about two-thirds the risk of the buy-and-hold investor.
You may want to compare the $719,947 total return on the Super Model in table 23 with the $219,564 return on the buy signals in table 21. That latter sum is included in the total of table 23 since it represents the appreciation on the buy signals seen in column 3 of that table. The remainder of the gain is made up by the compounding effect of all the dividends earned in column 4 plus all of the interest earned in column 6.
Table 24 shows how the Super Model worked when traded against the S&P 500 Index. An investor would have earned money in the buy signal periods twelve of the fourteen times, losing a small 5.3% in total on the December 26, 1980, buy signal and 3.2% on the February 7, 1989, buy signal. Of course, there would always be positive returns on interest during the sell periods. As seen at the bottom of the table, after twenty-seven years, $10,000 would have grown to $490,919, an annualized gain of 13.9%. Buy-and-hold on the S&P 500, including dividends, would have turned $10,000 into $222,816, a per annum return of only 10%. So, even on the farless-volatile S&P 500, an investor would have beaten that benchmark by 4 percentage points a year over a thirty-year stretch, and would have an ending portfolio more than two times as large in dollars as would the buy-and-hold investor.
TABLE 24
TOTAL RETURNS ON SUPER MODEL VS. STANDARD & POOR’S 500 INDEX: 1966 to 1996
Date
Signal
% Appreciation on Buys
Estimated Dividents on Buys
Total Return on Buys
Internet Earned in Sell Periods
$10,000 Growth
3/10/66
Sell +3.2 $10,320
11/16/66
Buy +19.4 +6.1 +25.5 12,952
7/26/68
Sell +.9 13,068
8/30/68
Buy +5.1 +1.0 +6.1 13,865
12/31/68
Sell +9.8 15,224
5/29/70
Buy +29.5 +4.9 +34.4 20,461
7/16/71
Sell +1.4 20,748
11/19/71
Buy +17.3 +2.1 +19.4 24,733
6/26/72
Sell +10.0 27,250
2/14/74
Buy +3.3 +.5 +3.8 28,286
3/29/74
Sell +4.2 29,473
10/15/74
Buy +37.1 +14.0 +51.1 44,534
5/31/77
Sell +27.1 56,603
5/6/80
Buy +21.6 +3.2 +24.8 70,641
12/12/80
Sell +.7 71,135
12/26/80
Buy -5.3 +2.4 -2.9 69,072
7/10/81
Sell +3.1 71,214
9/21/81
Buy +41.5 +12.0 +53.5 109,313
10/21/83
Sell +9.2 119,370
10/15/84
Buy +89.9 +12.2 +102.1 241,247
9/18/87
Sell +1.6 245,107
12/18/87
Buy +1.5 +1.6 +3.1 252,705
5/20/88
Sell +3.6 261,802
2/7/89
Buy -3.2 +.4 -2.8 254,472
3/20/89
Sell +3.5 263,378
8/24/89
Buy +27.8 +19.9 +47.7 389,010
5/17/94
Sell +5.87 411,845
7/6/95
Buy +17.3 +1.9 +19.2 490,919
3/20/96 *
Total Super Model re
turn: $490,919
Annualized return on Super Model: +13.9%
Buy-and-hold on S&P 500(including dividends): $222,816
Annualized return on buy-and-hold: $10.9%
These tests on the Super Model, recall, use a buy rule of 6 points and a sell rule of 3 points on a scale from zero to 10. You could modify the trading rules, if you choose, so that the entire portfolio is not necessarily moved on any one signal. For example, you might be fully invested in stocks if the model were, say, 7 points or higher. If the model then fell to 5 or 6 points, you might sell off one-third of your portfolio, and remain two-thirds invested in stocks. If the Super Model then fell into the 3-to-4-point range, you might sell off a second third of the portfolio, leaving yourself one-third invested in stocks and two-thirds in Treasury bills. Finally, if the Super Model were to drop to 2 points or less, you could then go 100% into cash equivalents and be totally out of the stock market.
You could devise similar schemes for being, say, 0%, 50%, or 100% invested. Or you could get even more complicated and have ranges of 0%, 25%, 50%, 75%, and 100% invested. The point is, you don’t have to move from 0% invested to 100% invested in one fell swoop. You should use the model in the way that makes you most comfortable. Remember, the odds are best for the market when the model is highest, and worst when the model is lowest.