#1)  Your historical results are for groups of 100 or more stocks that are bought and sold in a month or a quarter.  How is the ordinary investor supposed to duplicate these results?

If you assume that our approach has some validity, then using our data should improve the odds of a gain even on a single stock purchase.  In fact, it's just as likely that you'll beat the performance of the larger group on a single stock purchase as it is that you'll underperform the larger group.  It's kind of like holding a partially "loaded" coin...you can lose on any particular throw, but would you rather throw that coin or an ordinary one?  

2)  If this approach is so wonderful, why are you publicizing it at all?

Because we're not rich.  However, you can bet that a good chunk of our advertising revenues will get invested in the stock market using our own methods.

Ultimately, we think our experiments in using the historical data to turn a profit speak for themselves.  In virtually every case we beat the market substantially, given a reasonable timeframe and an occasional setback  

Note that most of these experiments involve investment periods of a month or a quarter, so commissions and bid/ask spreads should be minimal compared to the percentage gains gotten.  We're not pushing day-trading strategies that work fantastically until you add up commissions, price spreads, and tax hassles.

3)  Your historical experiments have interesting results, but aren't you cheating?  Ordinarily, an investment newsletter puts out a model portfolio and subscribers follow it over the years, making it difficult to exaggerate gains.  The start date on your "experiments", however, predates the web site by several years.  It's like saying, "I would have bought Microsoft in 1985, so I'll include it in my model portfolio."

The difference is, our stock selection process was, is, and will be purely mechanical.  We scan historical data for trends, and use those trends to make future predictions.  There's no element of subjectivity here...the computer does the "selecting".

4)  OK, I don't doubt your honesty or the results in your "experiments".  Can't you simplify the website and just give viewers a list of promising stocks instead of this maze of numbers?

Despite all the numbers, it shouldn't be horribly difficult to make your own list.  Check out the historical data.  It's up to you which tables of data you wish to use...quarterly, monthly, dual, last year's versus data compounded over several years, etc.  Then download the most current data and screen it according to the criteria you've settled on.

We don't envision the website as a standard recommendation service...there are plenty of those.  The functions of the site, we think, are spelled out reasonably clearly in our "spiel".  The freedom to choose amongst various approaches should not be considered a disadvantage.

5)  What are some of the potential pitfalls in using this historical data to make stock picks?

There are a lot.

For example, we note that over the period 1998-2003, a great way to invest was simply to buy cheap stocks, sell at a higher price, and buy yet more cheap stocks.  Obviously, if this focus on cheap stocks is merely a long term cycle, and it reverses next month, you could lose a lot of money by buying cheap stocks.

The whole web site operates under the assumption that history (short to mid-term history in particular, since we usually don't look back more than 25 years) repeats, to some extent.  If it doesn't, or it reverses, our historical data is less than useless.

Look at our investment tests...while we've outperformed the market over the long haul, there certainly have been periods where following our historical data resulted in losses that exceeded those of the general market.

Hopefully, our "risk-adjusted" data can mitigate some of the dangers of a major reversal in the trends we look at.

6)  How "clean" is your data?

The main source of "dirt" would be stock splits.  Regarding historical data, we adjust for splits as soon as we can.  Even if we miss an occasional split, we've written our analysis program so that large drops in expensive stocks or big gains in cheaper stocks (a possible reverse split) cause us to remove those particular stocks as sources of historical data.  The above filter also removes stocks with one-time large dividends, which can have the same effect as splits.

The above filter itself might be considered a source of dirt, but when we relax or strengthen our screening criteria, our historical data tables don't change much.  Apparently, given the large volume of data we test, it's difficult to skew the historical results to any great extent.

There is also the issue of "survivorship".  The stocks from which we extract our older data tend to be more stable, less volatile, and lower-tech, than more recent stocks.

7) Are you statisticians?

No!

The heart of our analysis is a relatively straightforward Monte-Carlo approach. Anytime we perform an operation (add, multiply, whatever) on a gain or loss that is associated with a particular stock, we also do the operation on another gain or loss that was randomly selected from the entire body of gains and losses. We save the best and worst randomly-derived results and compare them against the results of following various strategies. When we get an average gain associated with buying, say, the most expensive 10% of stocks, we're also calculating a randomly-derived gain (actually, many, which are later averaged) alongside it.

Even when we do our "normal distribution" calculations, these randomly-derived high and low values are the basis for comparison against our strategic high and low values.

It's also easy to calculate probabilities with the Monte Carlo approach. If you look at 100 different strategies and, on average, it takes 100 different Monte Carlo runs to arrive at a gain that exceeds that of the single best strategy, you can say there's a 50% chance that your best strategy was simply the result of luck and you haven't identified any patterns that are worth paying special attention to.

We feel the approach is fairly robust. Just a few years ago, limitations in processing power made such calculations impractical, but now we don't see any compelling reason to take any other statistical route. If you are an academic and wish to query us, we're open to discussion, but please don't assume we'll understand every statistical term you throw at us. Also, we have no intention of publishing our work in a journal, so we may or may not be inclined to discuss or justify our algorithms.

8)  Tell me about your proprietary indicators.

They're proprietary!  We do give some clues as to what they're all about on our "spiel" page.  They involve stock/stock correlations...correlating the performance of one stock to the performances of others.  Besides our desire to keep these indicators secret, the algorithms are also a bit complex and we occasionally attempt to tweak them to improve results.

9)  OK, I've referenced current data against historical data and have a list of 200 possible buys.  What now?

If you're a massive institution, buy them all.

More realistically, though, we can't say with certainty.  Our intuition is to use some sort of complementary technique to screen the list further...use fundamentals, industry trends, charts, whatever.  Another approach would be to look at the list of buys for stocks you're familiar and comfortable with.  The reason we don't attempt to narrow our buy possibilities down to, say, 10 stocks, is because we lose statistical significance in the process.  That is, further screening on our part could actually be counterproductive, so it's up to you to take the next step.

Of course, our list of possible buys isn't necessarily everyone's "starting point".  Maybe you already have a list of 20 fundamentally sound stocks, and you're looking for a way to whittle down the list even further.  We think referencing our current data against our historical data would be a great way to do that whittling.

In cases where seasonality favors particular industry groups, of course, it's possible to purchase ETF's or no-load mutual funds instead of individual stocks.

10)  How do I use "dual" data to go about screening stocks?

Let's say the historical data says a good strategy for the time period you're interested in is to buy the cheapest 20% of stocks combined with the highest 20% of "3monthgain".  First you take the current data and delete the 80% most expensive stocks using whatever database program you like.  If you started with 2000 lines of data, now you'll have 400 (2000 * 0.2).  Then you sort those 400 stocks according to the top 20% of "3monthgain".  Now you'll have a list of 80 (400 * 0.2) prime candidates.

The question arises as to whether you could, in the above case, first sort via "3monthgain" and then via "recent price" (as opposed to first using "recent price" and then "3monthgain").  This is a somewhat complex question...the answer would be "yes" if all columns of indicators were unrelated to each other (in statistics, no "multicollinearity").  However, there certainly are related columns in our data...a stock with a low "1monthgain" will most likely have a low "3monthgain".  On the other hand, we screen out highly multicollinear combinations of data in our analysis program, so the combinations of indicators you see in our "dual" historical data shouldn't be strongly related.  To make things as simple as possible, it's advisable to sort the data in the order suggested, though it probably doesn't make much difference.

Seasonality Service home page

Copyright © 2008 MarketSynopsis.com. All rights reserved.