The Fifth Week: June 12 - June 16

The fifth week was a fun one. I spent some time just cleaning up the code I had been working with so we could send our new stuff to IBM. Throughout that process I made some subtle adjustments to some of the algorithms, and some new observations as well.

There are a couple of new algorithms listed in the results below. Both AGH and ARG are simpler versions of the ZIP/CLIFF idea. ARG is simpler than AGH.

We investigated three different ways of dealing with old orders. One way was to have old orders expire after a certain amount of time. Another was to allow agents to revise their old orders if they wanted to. And a third was to allow agents to revise, but to expire their orders if they took too long to revise. We got some interesting results:

Results with the expiration policy but NO revision option:

Algorithm	Avg. Efficiency*	Trade Ratio**
1. GDS(mem=1)	0.997633	0.964333
2. GDL(mem=1)	0.994967	0.958167
3. AGH	0.993367	0.982000
4. ARG	0.992100	0.987833
5. ZIP	0.990900	0.981833
6. CLIFF	0.989233	0.980167
7. PVT	0.973067	0.964500
8. GDT(mem=1)	0.963900	0.952333
9. ZIC	0.913000	0.864833

Results with the expiration policy AND the revision option:

Algorithm	Avg. Efficiency*	Trade Ratio**
1. AGH	0.996067	0.991333
2. CLIFF	0.994567	0.994167
3. ZIP	0.994133	0.994833
4. ARG	0.994100	0.996333
5. PVT	0.992133	0.999167
6. GDL(mem=1)	0.986833	0.951500
7. GDS(mem=1)	0.982600	0.939167
8. ZIC	0.947967	0.944833
9. GDT(mem=1)	0.916700	0.883833

Results with the revision option but NO expiration policy:

Algorithm	Avg. Efficiency*	Trade Ratio**
1. AGH	0.995300	0.994333
2. CLIFF	0.995233	0.996167
3. ARG	0.994933	0.997333
4. ZIP	0.994900	0.995167
5. PVT	0.992667	0.997667
6. ZIC	0.951700	0.947167
7. GDL(mem=1)	0.779233	0.689333
8. GDS(mem=1)	0.665667	0.585333
9. GDT(mem=1)	0.660800	0.613500

* - Average efficiency is the ratio of actual surplus to theoretical surplus.

** - Trade ratio is the ratio of the actual number of trades to the theoretical number of trades.