Market Indicator Insight

Month... I turned 60 years old in 2006 and have been investing in stocks since I was 22 years old. For the last 15 years or so I've been hearing the saying, "Sell in May and go away." Supposedly the market goes no where at the end of May. Could this be a market indicator? I figured I better check this out and discovered it was somewhat true. See the table below, the data shows 424 seven day S&P 500 cycles started in the month of December (Month 12). To get you oriented... my database goes back approximately 20 years, there are roughly 21 trading days in December, there is a new seven day cycle started each trading day... 21 x 20 = 420. Per the table, the 424 seven day cycles averaged a .59% gain. December is the best month to buy the S&P 500 and hold for seven market days. Interestingly, May is the second best month to buy. This doesn't seem to agree with old saying "Sell in May" though. Hmm, maybe it's because the guy who made up the saying wasn't buying for seven market days and then selling. He was probably buying and holding through July, August and September. These are clearly the worst months to buy into a seven day cycle. Why is that? Well I'd guess that a lot of people get Christmas bonuses and generally invest more during the winter months. Possibly they sell stock during the summer to buy houses, cars or go on vacations. More houses sell when the weather is warmer. Also people do more driving during the summer. Why not buy a new car then instead of during the winter? At any rate, the Stock-Optimizer neural network now knows all about monthly cycles.

Avg % Profit by Month

Occurrences	Month	% S&P500 Change Over 7 Market Days
424	12	0.590508353099099
423	5	0.575446570999979
408	11	0.525395824130893
416	1	0.521911524466771
444	10	0.417720748823939
411	4	0.338145804514441
382	2	0.313474744035706
440	3	0.193393231313991
429	6	0.1514878477988
439	8	-.00632163660843717
421	7	-.0436749692957062
402	9	-.0610386564264619

Date... Over the years I've also heard that mutual funds tend to do a lot of buying around the end of each month. This doesn't mean too much if you buy a stock and hold it for a year. But look below, the 25th of the month is a super time to buy into a seven day cycle. For that matter, any day around the end of the month is a great time to buy. Look down at the bottom of the table, the 12th, 15th and 16th are terrible days to buy into a seven day cycle. Since these are losing entry days on average maybe we should do some selling on these days. That way, at the end of the seven day cycle we won't have lost as much money. Looks like those mutual fund guys really do start buying at the end of the month. Who knows, maybe they get lots of 401k money in during the month and then they have to invest it by the end of the month before the next month's money starts rolling in?? I don't work at a mutual fund company so I don't know. It doesn't really matter though, the neural network has a calendar and it knows what day of the month it is.

Avg % Profit by Date

Occurrences	Day	% S&P500 Change Over 7 Market Days
149	25	0.825799167655455
163	26	0.798723516847481
165	24	0.766003028118324
168	23	0.721641117853049
170	22	0.665137470939716
152	30	0.662578108480236
165	27	0.56807243252993
171	10	0.566737299623946
169	9	0.516049452384446
168	20	0.417431882783727
166	21	0.411383424318333
154	29	0.400039243493034
97	31	0.357523466347431
167	11	0.331317667161533
165	28	0.328709477950801
169	6	0.288866752099117
166	19	0.27849196320234
169	7	0.203339712515215
153	1	0.156812937055189
152	4	0.148319612872678
166	18	0.09712859248126
169	14	.0782798847963473
169	13	.0754553565930583
164	5	.0603764149956989
165	17	.0337357733981044
172	8	-.0130779599350026
164	2	-.0265045206911458
167	3	-.0313464013172671
170	12	-0.123690041468212
168	15	-0.174686820689999
167	16	-0.205094121267653

Weekday... Concerning profitability, I've never heard anyone say anything about one day of the week being better than any other day of the week. However, I've been investing for many years now and I've never yet seen a big move up on a Friday. And, I've seen the market sell off big on Friday lots of times. I figured the other calendar stuff was pretty good information so I might as well check the day of the week out also. When I first looked at the data below it didn't seem like my poor Friday theory held water. However, I started thinking about it a bit more. Since we deal with a seven market day cycle sometimes there will be two Fridays in a cycle and sometimes there will only be one Friday in a cycle. It depends on which day of the week the prediction comes out. If the neural network prediction comes out on Monday night you buy on Tuesday morning. You then hold on Wednesday, Thursday, Friday, Monday, Tuesday, and Wednesday. That's seven market days, only one Friday is in this particular cycle. Note in the table below that predictions coming out on Monday were second best. What it boils down to is predictions coming out on Thursday and Wednesday have two Fridays in their cycles. The other days have just one. According to the data, Wednesday and Thursday are the worst days to receive predictions. Maybe my thinking wasn't so far off base after all. On the other hand, at first glance there doesn't seem to be a large advantage in this information. I did more analysis on the problem and found out a very interesting fact though. All Thursdays are not the same and all Wednesdays are not the same. Under certain circumstances Wednesdays and Thursdays are terrible, actually negative. A neural network picks up these circumstances, they are called interactions. Therefore, I included "Weekday" as an indicator. BTW - the reason why Fridays are often bad days is because short term traders are afraid that bad news might come out over the weekend. They like to be neutral over the weekend so they often sell on Friday, especially if the market isn't acting well. Maybe that's the circumstance I'm referring to above but I'm not telling. I know there will be many financial wizards looking at all this stuff trying to steal my ideas. If you are one of them don't bother emailing me with questions about indicators, I'm already blabbing way too much anyway... I must be nuts!!!

Avg % Profit by Weekday

Occurrences	Day	% S&P500 Change Over 7 Market Days
1032	Tuesday	0.332816891603792
957	Monday	0.319841284129083
1007	Friday	0.288536035485951
1012	Thursday	0.271824346932741
1031	Wednesday	0.250866791662822

TRIN function... I'm not going to try to mathematically explain what the TRIN is. Instead I'll provide this Google link. If you are inclined to delve into it you can find out more about it. What I will say is that I've developed a proprietary derivation of the TRIN. This function isn't an overbought/oversold type indicator. Based on recent market volume related to advancing and declining issues, it indicates likely market direction during the next seven market days. What it doesn't do is suggest how much up or down the market is likely to go. It just suggests direction but its one heck of a powerful indicator. I actually use two neural networks, the TRIN function tells me which one to use on any given night. One neural network specializes in analyzing high probability "Up" cycles (see table below) while the other specializes in analyzing high probability "Down" cycles.

TRIN function

Occurrences	Direction	% S&P500 Change Over 7 Market Days
4192	Up	.36698
803	Down	-.22155

Thus far I've have been discussing discrete indicators. Discrete indicators offer a finite number of input choices. There are only 5 inputs choices in the "Weekday" indicator. Since the market is closed over the weekend the input has to be Monday, Tuesday, Wednesday, Thursday or Friday. Continuous indicators offer an infinite number of input choices. An example of a continuous indicator would be any number between 1 and 10. There are an infinite choices, 3.56439 is one choice. Below I will discuss the continuous indicators a bit but I can't show tables on these because of their nature. Instead I'll speak in terms of correlation and overbought/oversold conditions.

There are a couple of old sayings, "Trees don't grow to the sky" and "You can't dig a hole to China." Both of these sayings are very relevant to the stock market. While the market can at times become very overbought or very oversold, on average it reverts to its mean. I didn't make this part up, nor did I hear it from anyone. It's "just the fact Jack", the historical data is irrefutable. The market very rarely goes up day after day without pause. It tends to stop and backfill a bit and then possibly continues the uptrend. The indicators discussed below tell the neural network how overbought or oversold the market currently is. It can then compare current market conditions with similar past conditions and relate them to past outcomes. It essentially bases its prediction on these past outcomes. If history repeats itself the predictions come out great. Why would history repeat itself? Well, trees don't grow to the sky and nobody ever hit China yet, it's about the same thing.

T10-14... % of stocks one sigma above their 40 day moving average minus % of stocks one sigma below their 40 day moving average. This indicator is a statistical measure of overbought/oversold conditions over the last 40 market days. There is a negative correlation. The higher T10-14 gets, the bigger the expected S&P seven day cycle loss. The lower T10-14 gets, the bigger the expected S&P 500 seven day cycle gain.

T11-15... % of stocks two sigma above their 200 day moving average minus % of stocks two sigma below their 200 day moving average. This indicator is a statistical measure of overbought/oversold conditions over the last 200 market days. There is a negative correlation. The higher T11-15 gets, the bigger the expected S&P seven day cycle loss. The lower T11-15 gets, the bigger the expected S&P 500 seven day cycle gain.

T12-16 function... A proprietary function of % of stocks two sigma above their 40 day moving average minus % of stocks two sigma below their 40 day moving average. This indicator is a statistical measure of overbought/oversold conditions over the last 40 market days. There is a positive correlation. The lower T12-16 function gets, the bigger the expected S&P seven day cycle loss. The higher T12-16 function gets, the bigger the expected S&P 500 seven day cycle gain.

C/C7... Close of S&P 500 divided by Close of S&P 500 seven market days ago. This is a very simple seven day momentum indicator. There is a negative correlation. The higher C/C7 gets, the bigger the expected S&P 500 seven day cycle loss. The lower C/C7 gets, the bigger the expected S&P 500 seven day cycle gain.

C/C1... Close of S&P 500 divided by Close of S&P 500 one market day ago. This is a very simple one day momentum indicator. There is a negative correlation. The higher C/C1 gets, the bigger the expected S&P 500 seven day cycle loss. The lower C/C1 gets, the bigger the expected S&P 500 seven day cycle gain.