Bot definition

I will define an address as a bot when for the last 30 days one daily count is bigger than the parameter max_txs_humans. This is achieved calculating the maximum daily txs per address in a CTE and joining it with the swap information from terra.swaps table. Note that the time frame is also a parameter, so the analysis can be extended to 60 and 90 days.

Bot vs human behaviour

The behaviour of both type of addresses will be compared with the following metrics:

• Number of swaps, daily and cumulative
• Swapped volume (USD) and fees volume (USD), daily and cumulative
• Percentage of fees over swapped volume

From Figures 1. to 3., a important difference in behaviour can be identified: bots make less trades but bigger volume than humans. Humans account for 200k swaps for the 74k from bots, but volume is 248k USD compared to over 2B. Scale is just totally different! There seems to be some missing data in the beginning of April for these high value transactions which bots are identified with (or maybe there is a bot holiday I don't know about and all bots were having some rest from their busy lives).

Percentage of swaps executed by bots in the last days

Figure 5 is a normalized view of the cumulative swaps for humans and bots. So the value in April 25th would answer the question what percentage of swaps were executed by bots in the past 30 days. 26.9% of swaps were bots compared to 73.1% humans

Figure 4. shows the fee in percentage of swapped volume by address type. There is a change in the trend around mid April when fee percentage by bots is bigger then human. It can also be appreciated in the relation between bot curves in Figures 2. and 3. compared to the flatter profiles for humans. All in all it is a very small fee in any case (below 0.2% for the most part).

Figure 6. shows the number of swaps for each address type by day of the week and hour of the day. Is an attempt to be creative with visualizations and show seasonal relationships. Although the readibility of the graphs could be improved, it shows very clear patterns in human behaviour, concentrated from noon til late afternoon (UTC) and on Sundays and Mondays. Bot patterns look more random at first glance, but this promising visualization deserves a closer look to improve the information transmission.

Popular pairs among bot swappers

Figure 7. shows the top 10 swaps pairs for bots, with a log scale on the number of swaps. The LUNA-UST pair is used in over 98.5% of the swaps (72.9k over 74k).

Extra Insight - Bot swaps vs LUNA price correlation

Since the analysis of the popular pairs among bot swappers led to the fact that almost all swaps are LUNA-UST, I decided to confirm the correlation between the LUNA price and bot activity.

Figure 8. shows the positive correlation between LUNA price and bot swapping activity.

Conclusion

Bot addresses swap less times, more volume and almost exclusively on LUNA-UST pair as opposed to humans, which make almost 3 times more swaps but one tenth of the volume.

Also, humans show clear seasonal habits as opposed to random activity of bots.