Thoughts on the State of the Market (WE 01/24/2009)
The State of the Market report is looking for a bearish start to the trading week, but the confidence in that call is smaller than usual because of the “abnormal market filter” I wrote about on Saturday.
Trading strategy MarketSci will start the week neutral, and YK and Scotty both net short (with position sizes reduced as a result of the same abnormal filter).
Taken as a whole, I’m bearish but cautious for Monday.
Seasonality side note – from Tuesday on we’ll be in the turn of the month, a historically bullish period in the stock market.
What’s Working Right Now
To the uninitiated, the State of the Market report is a free snapshot of what many of the strategies I’ve tested on this blog are saying about the stock market right now.
The report correctly called 4 out of 4 closing directions this week, although the confidence of this week’s predictions has been especially small.
For those keeping score at home, since the launch of the SOTM back in November of 2008, the report has called 71% of all days correctly with winners 0.8x losers (I’ve multiplied each day’s % change by the SOTM’s aggregate prediction to capture not just the direction of the prediction, but also the confidence).
The most effective SOTM strategy of the last couple of weeks has been adaptive daily follow-through (7 of 9 days correct with winning predictions 1.5x losers), extreme RSI(2) (4 of 5 days correct, winners 0.9x the loser), and the proprietary VIX-based indicator (8 of 9 days correct, winners 0.4x the loser).
The furthest off the reservation continues to be the SOXX leadership strategy (3 of 9 days correct, winners 0.6x losers). This strategy has been performing so poorly for so long that the adaptive nature of the SOTM report has all but removed its influence on the aggregate prediction, but for the time being, I’ll keep it on the report. I think it’s a good opportunity for readers to see the report adapt in real-time.
Happy Trading,
ms
To stay up to date with what’s happening at the MarketSci Blog, we recommend subscribing to our RSS Feed or Email Feed.
Filed under: State of the Market Report | 13 Comments
Michael,
first of all I beg your pardon for not being a native speaker, but I really appreciate your work.
Concerning your daily “state of the market” report there seems to be another seasonality (“short term”) which may be worth some considerations.
At least for the last 2 years there is a clear tendency of the equity market(s) to start the week weak (on Mondays) and close the week on a strong note (Thursday and/or Fridays).
The distribution of up and down days for the different days of the week is as follows (S&P 500, closing values):
last 63 trading days (3-month period):
Mondays Up-days / Down-Days: 2/11
Tuesdays Up-days / Down-Days: 9/3
Wednesdays Up-days / Down-Days: 6/7
Thursdays Up-days / Down-Days: 7/6
Fridays Up-days / Down-Days: 8/3
last 252 trading days (12-month period):
Mondays Up-days / Down-Days: 19/31
Tuesdays Up-days / Down-Days: 26/22
Wednesdays Up-days / Down-Days: 21/31
Thursdays Up-days / Down-Days: 34/18
Fridays Up-days / Down-Days: 28/21
since 01/02/1990:
Mondays Up-days / Down-Days: 523/436
Tuesdays Up-days / Down-Days: 461/449
Wednesdays Up-days / Down-Days: 529/454
Thursdays Up-days / Down-Days: 512/469
Fridays Up-days / Down-Days: 507/458
Looking back the last 5 years up days and down days are more evenly distributed, but at least the odds are in favor or 1.4:1 for an up day on Thursday and 1.2:1 for an up day on Friday.
These figures seems to tilt the odds in favor of going short on Friday’s close (exit the trade on Monday’s close) and going long on Wednesday’s or Thursday’s close (exit the trade on Friday’s close), but the results concerning the implementation of a simple trading system would be far less impressive because the system would have been in the market at the worst possible moments like going short on Friday, October 10 2008 with a loss of -11.58% the next day (a Monday, the S&P 500 gained +11.58%)
During the last 252 trading days only the short side would have been profitable with a win/loss ratio of 1.73, a percentage of winning trades of 62.00% and an average winning trade of +2.00% / an average loosing trade of -1.89%, but still wouldn’t have outperformed a simple “buy and hold” strategy (means: going short a year ago).
But although the tradability on its own is more than questionable, it seems worth to be taken into consideration for timing one’s investments (as long as the odds remain unchanged).
Best regards
Frank
RE to Frank: good stuff sir. I’ll run some numbers and possibly do a post on this in the near future. All the best, michael
ms,
You mention that due to the “adaptive nature” of the report, the weighting of SOXX has/is reduced. Thus the current influence is minimal.
Would it not be an idea to keep the strategy @ weighting X, [and due to various strategies coming in/going out of fashion over time] you may find that in a future time period it’s weighting starts to rise, as that strategy regains whatever original edge it must have had at some point.
As you gather a large number of individual strategies, some working well, some not so well, you should be able to discern [if there exists one] a macro-strategy, composed of similar backtested edges, viz. trend following [bull/bear market] reversion [consolidating market, or accumulation/distribution, possibly indicating major turning points]
Interesting stuff.
jog on
duc
Michael,
thanks for the flowers.
But there some more regularities and/or abnormalities concerning the last trading day of the week (regularly a Friday):
What about not going short the S&P 500 on Friday’s close no matter what but only on Fridays which are closing up ?
Looking 2 days ahead for a lower close (than Friday’s close) the first or the second session later (regularly Monday or Tuesday) in comparison to the at-any-time odds for a lower close during the next 2 sessions, the (impressive) figures are:
The S&P 500 closed lower (than Friday) the first or the second session later (regularly Monday or Tuesday):
-last 252 trading days ( 1 year ): 29 occurrences: 26 winner/3 looser (89.66%) vs. at-any-time odds of 68.00% for a lower close during the next 2 sessions
-last 504 trading days ( 2 years): 55 occurrences: 47 winner/8 looser (85.45%) vs. at-any-time odds of 65.14% for a lower close during the next 2 sessions
-last 756 trading days ( 3 years): 81 occurrences: 69 winner/12 looser (85.19%) vs. at-any-time odds of 63.13% for a lower close during the next 2 sessions
-last 1008 trading days ( 4 years): 111 occurrences: 89 winner/22 looser (80.18%) vs. at-any-time odds of 62.33% for a lower close during the next 2 sessions
But what about not going short the S&P 500 on up-Friday’s but on those Friday’s only when the S&P 500 closed up the session before as well (up-days on Thursday and Friday) ?
The S&P 500 closed lower (than Friday) the first or the second session later (regularly Monday or Tuesday):
-last 252 trading days ( 1 year ): 19 occurrences: 19 winner/0 looser (100%) vs. at-any-time odds of 68.00% for a lower close during the next 2 sessions
-last 504 trading days ( 2 years): 32 occurrences: 29 winner/3 looser (90.63%) vs. at-any-time odds of 65.14% for a lower close during the next 2 sessions
-last 756 trading days ( 3 years): 44 occurrences: 40 winner/4 looser (90.91%) vs. at-any-time odds of 63.13% for a lower close during the next 2 sessions
-last 1008 trading days ( 4 years): 61 occurrences: 52 winner/9 looser (85.25%) vs. at-any-time odds of 62.33% for a lower close during the next 2 sessions
In both studies the probabilities remain constant over time, so they seem not correlated to the market environment (bull or bear markets).
No other trading day of the week produced similar results in comparison to the at-any-time odds, neither on the short nor on the long side.
The last trading day of the week seems to be a session when going short the S&P 500 could produce outstanding results. If wanted I’ll come up with some other figures (Profit Factor, Avg. Profit per Trade, Avg. Loss per Trade) a bit later.
Disclaimer: This analysis has been prepared to the best of my knowledge. There is no guarantee that the data and analysis is correct, everybody has to do his own independently research.
Best regards
Frank
RE to duc: so if I understand you correctly, you’re suggesting that strategy X has some “core value”. If it slips below that core value enough, it’s weighting gets reduced, but if it begins to perform well again, it jumps back up to its core value. Do I have that right? Interesting idea…let me ponder a bit.
RE macro strategy – I hope that’s basically what we’ve created with the “aggregate prediction”. I chose to add the particular strategies that I did because they have all been historically effective, but are using completely different data sources or approaches to the market to come to their own conclusions. The aggregate prediction is an attempt to combine those in a smart way into one all encompassing prediction. This is still a pretty new concept. Since its inception in November, we’ve been on the right side of the market about 71% of the time with winning predictions 0.8x losing ones. Of course, today is not the best example of that effectiveness =|
jog on sir
RE to Frank: in case this turns into a post, where are you from? Andrey from Russia is a big poster here. I just like to know where my contributors hail from (especially when they’re international). Thanks, michael
Michael,
I’m from Germany.
Best
Frank
ms,
In essence yes. In the same way a moving average weights the most recent data, so you could weight the signals from the various strategies that you are running, based on their performance.
In this way the “group” of strategies, become a meta-strategy. They become as a meta-strategy [to an extent] adaptive to the market. They would also indicate to a degree [possibly] the “style” of trading, for example trend following, that was successful at the current point in time.
jog on
duc
RE to duc: in a roundabout way that’s what we’re doing with an extra small (but important) step – multiple strategies could be performing well at some given time that are basically iterations of each other (intermediate-term contrarian for instance). I’ve reduced weighting in this case manually (it could be done mathematically to your point, but i haven’t gone that far) and increased the weighting of strategies that are taking particularly novel approaches to the market.
I think that this aggregate approach will key in on the style of trading that is prevailing at a very high level at any given time. By that I mean if something works in some given month, the strategy isn’t going to radically rework itself. As the market evolves in a big way (as we’ve seen with daily follow-through for instance) the aggregate prediction should pick up on that.
Good stuff sir. jog on
ms,
Yes, I see. Clever.
jog on
duc
Ahhh…the age-old system trader’s dilemma….when to turn on the system, when to turn it off…When to allocate extra capital to it, when it underweight it…
I hope to be able to provide some clarity (quantifiable) on that over the next couple of months.
And the ramifications are huge..If there was a way to determine, without ambiguity, that a system was breaking down, what would that mean? Well it could mean that we could trade ALL the recently over-performing systems, snatching huge gains from the markets, with little risk that they’ll quit working before we are aware that they have quit working…
For frank,
Your strategy does perform well from 2004 to today but completely blew up from 1986 to 2004.
Shivar
Shiver,
you’d be totally correct if one would have played the same strategy (going short the S&P 500 on close of Friday if Friday and the session before would have been up days), but this kind of strategy heavily depends on the then current distribution of up and down days concerning the days of the week.
Between 1990 and the end of 2004 the distribution would have been totally different to the environment experienced during the last four years. Between 1990 and the end of 2004 Fridays performed the worse (in opposite to the current environment) of all weekdays (399 up-days vs. 361 down-days), while Wednesday was the best performing weekday (445 up-days vs. 329 down-days), followed by Mondays -again in opposite to the current environment- with 406 up-days vs. 348 down-days.
Concerning this strategy you’d periodically have to look for some adaptive measurements -if any- if, when and to what extent the distribution of up and down days concerning the days of the week changes over time.
Best regards
Frank