Hi, professor Stoikov. I'm Yubo, a rising junior student, and this summer I'm studying your paper, The micro-price: a high-frequency estimator of future prices. When I work on the dataset, I find a problem that when I choose the different number of imbalance bucket, the behavior of micro adjustment conditional on imb_bucket would be changed unexpected, especially for setting more buckets (make sure micro-adj is still converged). For example, the first pic uses the dataset of CVX when spread is 4 ticksize, and we can see the difference when n_imb=10 and n_imb=4. Moreover, this difference become more obvious when I apply the method on the dataset of China Treasury Bond Futures Market, which is shown in pic2.
Obviously, the negative correlation around the imbalance=0.5 is not expected for the both. I'm just wondering is there any standard principle to choose the number of imbalance bucket? And, why does this phenomenon appear? Please give me some advice and I'm looking for your reply :)
Hi, professor Stoikov. I'm Yubo, a rising junior student, and this summer I'm studying your paper, The micro-price: a high-frequency estimator of future prices. When I work on the dataset, I find a problem that when I choose the different number of imbalance bucket, the behavior of micro adjustment conditional on imb_bucket would be changed unexpected, especially for setting more buckets (make sure micro-adj is still converged). For example, the first pic uses the dataset of CVX when spread is 4 ticksize, and we can see the difference when n_imb=10 and n_imb=4. Moreover, this difference become more obvious when I apply the method on the dataset of China Treasury Bond Futures Market, which is shown in pic2.
Obviously, the negative correlation around the imbalance=0.5 is not expected for the both. I'm just wondering is there any standard principle to choose the number of imbalance bucket? And, why does this phenomenon appear? Please give me some advice and I'm looking for your reply :)