Spatiotemporal localisation patterns of technological startups: the case for recurrent neural networks in predicting urban startup clusters

Abstract

AbstractMore attention should be dedicated to intra-urban localisation decisions of technological startups. While the general trend of innovative companies being attracted to metropolitan areas is well-known and thoroughly researched, much less is understood about the micro-geographical patterns emerging within cities. Considering the growing number of papers mentioning that agglomeration externalities attenuate sharply with distance, such an analysis of micro-scale localisation patterns is crucial for understanding whether these effects are of importance for technological startups. Using a sample of startups from the up-and-coming market in Central-East Europe in Warsaw, Poland, their spatial organisation across the years will be tracked to investigate whether there is a defined pattern consistent with highly localised externalities operating within cities and how this pattern evolves over time. Additionally, the paper will show how recurrent neural networks may help predict the locations of technological startup clusters. It will be presented how to include the spatial dimension in the model in a computationally effective way and how this augmentation improves the results by allowing the network to “understand” the spatial relations between neighbouring observations.