Measures used:
Search Volume = No. of searches in September 2006 for a few keywords
(obtained from Yahoo Keyword Selector).

Ordering by Search Volumes = Ranks of the words when arranged in
ascending order of search volume.

For the biddings, I consider their maximum, sum(of upto 40 top
biddings), mean and median.


EXPERIMENT I:
----------------

I use "car" as the main keyword, and take about the "top" 25-30 of the
"related" keywords:

Considering the Max Bid:

Google's sorting by relevance of its own suggestions gave poor
results: the correlation between Google's ordering and maximum
Overture bidding was about 0.2 . I left Google with this debacle.

Using the ordering given by Yahoo Search Volumes, it rose to 0.3402 .
When I used the actual search volume (instead of the ordering),
the natural thing(I feel) was for correlation to increase. So it did,
but by a very small amount, to 0.3647 . (But as we will see, the
situation changes very soon!)

Considering the sum of all bids:
It falls to 0.2689 for Yahoo Search volumes. For the ordering, it is
0.3430 .

Considering the mean of all bids: Doesn't help (0.2588,0.3244).

Considering the median of all bids: Doesn't help either(0.2437,0.2938).

Putting this in a table (only the Yahoo stuff, I forget about Google),
we get:

Correlation:

         Search Volume                Ordering by search volume

Maximum   0.3647                         0.3402

Sum       0.2689                         0.3430

Mean      0.2588                         0.3244

Median    0.2437                         0.2938

---------------------------------------------------------------------------

All these definitely tend to show that there is not much correlation
between search volumes and bidding values. I think it is an expected
result, because rarer keywords often reflect certain characteristics
of the prospective buyer, and such characteristics can be of high
value to a seller.

A trivial example: the word "hi" was searched 174965 times (on Yahoo,
in September 2006). This indicates a high search volume.
However, it has only 2 bidders, at $0.16 and $0.10 .


So, what measure, instead of search volume, can produce a better
correlation?


EXPERIMENT II:
--------------------

Another example: "insurance"

34 samples from Yahoo keyword selector.

Correlation:

         Search Volume                Ordering by search volume

Maximum   0.3665                         0.3674

Sum       0.5005                         0.5331

Mean      0.5062                         0.5197

Median    0.4058                         0.4914


With the word insurance itself having a significant indication(more
significant than the word car) about the intention of the user, the
correlation increases.

--------------------------------------------------------------------------

So, there can be higher correlation with search volume, but only if
the core word carries significant information about the user. More
experiments need to be carried out before building confidence on this
statement, but it seems natural.

The only unnatural thing that I find happening is the following:
When using the ranks of the words (when arranged in ascending order of
their search volumes), the correlation with bidding increases. Thus
the magnitude of difference in search volumes is not influencing the
market as much as it should. It can, of course, be a coincidence for
the two cases studied.